In [1]:
# This Python 3 environment comes with many helpful analytics libraries installed
# It is defined by the kaggle/python Docker image: https://github.com/kaggle/docker-python
# For example, here's several helpful packages to load

import numpy as np # linear algebra
import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)

# Input data files are available in the read-only "../input/" directory
# For example, running this (by clicking run or pressing Shift+Enter) will list all files under the input directory

import os
for dirname, _, filenames in os.walk('/kaggle/input'):
    for filename in filenames:
        if ".txt" in filename:
            print(os.path.join(dirname, filename))

# You can write up to 20GB to the current directory (/kaggle/working/) that gets preserved as output when you create a version using "Save & Run All" 
# You can also write temporary files to /kaggle/temp/, but they won't be saved outside of the current session

/kaggle/input/home-data/Houses Dataset/HousesInfo.txt

Import Statement¶

In [2]:
import pandas as pd
import numpy as np
import os 
import matplotlib.pyplot as plt

from collections.abc import Sequence
from sklearn import preprocessing
from sklearn.model_selection import train_test_split
from sklearn import metrics
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Activation, Dropout, Conv2D,MaxPooling2D, Flatten
from tensorflow.keras.callbacks import EarlyStopping
from tensorflow.keras.callbacks import ModelCheckpoint
In [3]:
import seaborn as sns
import csv
import numpy as np
%matplotlib inline

import cv2
import glob
from IPython.display import Image
In [4]:
import tensorflow as tf
tf.test.gpu_device_name()
Out[4]:
'/device:GPU:0'
In [5]:
def encode_text_dummy(df, name):
    dummies = pd.get_dummies(df[name])
    for x in dummies.columns:
        dummy_name = "{}-{}".format(name, x)
        df[dummy_name] = dummies[x]
    df.drop(name, axis=1, inplace=True)
    
def encode_numeric_zscore(df, name, mean=None, sd=None):
    if mean is None:
        mean = df[name].mean()

    if sd is None:
        sd = df[name].std()

    df[name] = (df[name] - mean) / sd

def missing_median(df, name):
    med = df[name].median()
    df[name] = df[name].fillna(med)
    
def to_xy(df, target):
    result = []
    for x in df.columns:
        if x != target:
            result.append(x)
    # find out the type of the target column.
    target_type = df[target].dtypes
    target_type = target_type[0] if isinstance(target_type, Sequence) else target_type
    # Encode to int for classification, float otherwise. TensorFlow likes 32 bits.
    if target_type in (np.int64, np.int32):
        # Classification
        dummies = pd.get_dummies(df[target])
        return df[result].values.astype(np.float32), dummies.values.astype(np.float32)
    else:
        # Regression
        return df[result].values.astype(np.float32), df[target].values.astype(np.float32)
    
def remove_outliers(df, name, sd):
    drop_rows = df.index[(np.abs(df[name] - df[name].mean()) >= (sd * df[name].std()))]
    df.drop(drop_rows, axis=0, inplace=True)

def encode_numeric_range(df, name, normalized_low=-1, normalized_high=1,
                         data_low=None, data_high=None):
    if data_low is None:
        data_low = min(df[name])
        data_high = max(df[name])

    df[name] = ((df[name] - data_low) / (data_high - data_low)) \
               * (normalized_high - normalized_low) + normalized_low

def chart_regression(pred,y,sort=True):
    t = pd.DataFrame({'pred' : pred, 'y' : y.flatten()})
    if sort:
        t.sort_values(by=['y'],inplace=True)
    a = plt.plot(t['y'].tolist(),label='expected')
    b = plt.plot(t['pred'].tolist(),label='prediction')
    plt.ylabel('output')
    plt.legend()
    plt.show()

Data Preprocessing¶

Import Data¶

In [6]:
filename_read = '/kaggle/input/home-data/Houses Dataset/HousesInfo.txt' 
# filename_read = './Houses Dataset/HousesInfo.txt'
In [7]:
cols=["Bedrooms","Bathrooms","area","zipcode","price"]
df = pd.read_csv(filename_read , sep=" ", header=None , names=cols)
In [8]:
df.head()
Out[8]:
Bedrooms Bathrooms area zipcode price
0 4 4.0 4053 85255 869500
1 4 3.0 3343 36372 865200
2 3 4.0 3923 85266 889000
3 5 5.0 4022 85262 910000
4 3 4.0 4116 85266 971226
In [9]:
df.shape
Out[9]:
(535, 5)
In [10]:
# Keep the house prices between 300k and 900k (all others maybe considered outliers)
# Show distribution

Handle Images dataset¶

In [11]:
filepath = "/kaggle/input/home-data/Houses Dataset/"
# filepath = './Houses Dataset/'
In [12]:
## Bathroom.jpg
new_images=[]
for number in range(1, df.shape[0] + 1):
    for path in glob.glob(filepath + str(number) + "_bathroom.jpg"):
        if os.path.isfile(path):
            new_images.append(path)
In [13]:
new_images[:5]
Out[13]:
['/kaggle/input/home-data/Houses Dataset/1_bathroom.jpg',
 '/kaggle/input/home-data/Houses Dataset/2_bathroom.jpg',
 '/kaggle/input/home-data/Houses Dataset/3_bathroom.jpg',
 '/kaggle/input/home-data/Houses Dataset/4_bathroom.jpg',
 '/kaggle/input/home-data/Houses Dataset/5_bathroom.jpg']
In [14]:
img= pd.DataFrame(new_images,columns = ['bathroom_img'])
In [15]:
## bedroom images
bedroom_images = []
for number in range(1, df.shape[0] + 1):
    for path in glob.glob(filepath + str(number) + "_bedroom.jpg"):
        if os.path.isfile(path):
            bedroom_images.append(path)
In [16]:
img['bedroom_img']=bedroom_images
In [17]:
frontal_images = []
for number in range(1, df.shape[0] + 1):
    for path in glob.glob(filepath + str(number) + "_frontal.jpg"):
        if os.path.isfile(path):
            frontal_images.append(path)
In [18]:
img['frontal_img']=frontal_images
In [19]:
kitchen_images = []
for number in range(1, df.shape[0] + 1):
    for path in glob.glob(filepath + str(number) + "_kitchen.jpg"):
        if os.path.isfile(path):
            kitchen_images.append(path)
In [20]:
img['kitchen_img']=kitchen_images
In [21]:
img.head()
Out[21]:
bathroom_img bedroom_img frontal_img kitchen_img
0 /kaggle/input/home-data/Houses Dataset/1_bathr... /kaggle/input/home-data/Houses Dataset/1_bedro... /kaggle/input/home-data/Houses Dataset/1_front... /kaggle/input/home-data/Houses Dataset/1_kitch...
1 /kaggle/input/home-data/Houses Dataset/2_bathr... /kaggle/input/home-data/Houses Dataset/2_bedro... /kaggle/input/home-data/Houses Dataset/2_front... /kaggle/input/home-data/Houses Dataset/2_kitch...
2 /kaggle/input/home-data/Houses Dataset/3_bathr... /kaggle/input/home-data/Houses Dataset/3_bedro... /kaggle/input/home-data/Houses Dataset/3_front... /kaggle/input/home-data/Houses Dataset/3_kitch...
3 /kaggle/input/home-data/Houses Dataset/4_bathr... /kaggle/input/home-data/Houses Dataset/4_bedro... /kaggle/input/home-data/Houses Dataset/4_front... /kaggle/input/home-data/Houses Dataset/4_kitch...
4 /kaggle/input/home-data/Houses Dataset/5_bathr... /kaggle/input/home-data/Houses Dataset/5_bedro... /kaggle/input/home-data/Houses Dataset/5_front... /kaggle/input/home-data/Houses Dataset/5_kitch...

Describe price distribution and handle outliers¶

In [22]:
price_summary = df['price'].describe()

price_summary
Out[22]:
count    5.350000e+02
mean     5.893628e+05
std      5.090261e+05
min      2.200000e+04
25%      2.492000e+05
50%      5.290000e+05
75%      7.285000e+05
max      5.858000e+06
Name: price, dtype: float64
In [23]:
# Print out the spread of prices using matplotlib
# Assuming df['prices'] contains your price data
plt.hist(df['price'], bins=30, edgecolor='black')
plt.title('Price Distribution')
plt.xlabel('Price')
plt.ylabel('Frequency')
plt.show()
No description has been provided for this image
In [24]:
# Remove values that are outliers
# df.iloc[row, col]
outliers = []
for i in range(0, len(df)):
    if df.iloc[i, :].price < 300000 or df.iloc[i, :].price > 900000: 
        outliers.append(i)
        # print("Outliers:", df.loc[outliers, 'price'])

#         print(i, df.iloc[i, :].price)

print("Outliers", len(outliers))
print("Remaining", 535 - len(outliers))

print(outliers)

Outliers 227
Remaining 308
[3, 4, 5, 7, 8, 9, 11, 18, 19, 20, 21, 23, 24, 25, 26, 28, 29, 31, 33, 34, 35, 36, 38, 50, 53, 58, 59, 60, 62, 63, 64, 65, 66, 67, 68, 70, 71, 72, 74, 75, 76, 77, 78, 79, 86, 90, 92, 94, 96, 102, 109, 111, 112, 115, 125, 128, 130, 132, 133, 136, 143, 146, 148, 149, 150, 151, 152, 153, 154, 155, 156, 157, 158, 160, 161, 164, 168, 170, 172, 184, 185, 186, 190, 192, 202, 203, 207, 209, 214, 215, 216, 218, 219, 220, 223, 224, 226, 252, 255, 256, 257, 260, 261, 263, 266, 268, 270, 285, 286, 287, 288, 289, 290, 291, 292, 293, 318, 320, 321, 330, 331, 332, 333, 334, 335, 336, 337, 338, 339, 340, 341, 342, 343, 344, 345, 346, 347, 348, 349, 350, 351, 352, 353, 354, 355, 356, 357, 358, 359, 360, 361, 362, 363, 364, 365, 366, 367, 368, 369, 370, 372, 373, 374, 375, 376, 377, 378, 379, 380, 381, 382, 383, 384, 385, 386, 387, 388, 389, 390, 391, 392, 393, 394, 395, 396, 397, 398, 399, 400, 401, 402, 403, 404, 405, 406, 407, 408, 409, 410, 411, 412, 413, 414, 415, 416, 417, 418, 419, 420, 421, 425, 457, 463, 465, 471, 472, 476, 478, 483, 484, 485, 486, 487, 491, 499, 506, 514]
In [25]:
data = df[~df.index.isin(outliers)]
In [26]:
# Print out the spread of prices using matplotlib
# Assuming df['prices'] contains your price data
plt.hist(data['price'], bins=30, edgecolor='black')
plt.title('Price Distribution')
plt.xlabel('Price')
plt.ylabel('Frequency')
plt.show()
No description has been provided for this image

Merge bathroom, bedroom, frontal, and kitchen into one image¶

In [27]:
images_output=[]
img_df = img
for row_index,row in img.iterrows():
    # Check if the row_index is in the outliers to do this do the outliers first
    if row_index in outliers: 
      img_df = img_df[img_df.index != row_index] # Remove outlier
      # print(img_df.shape)
      continue

    inputImages=[]
    outputImage = np.zeros((128, 128, 3), dtype="uint8")
    image_temp1 = cv2.imread(row.bathroom_img)
    image1 = cv2.resize(image_temp1, (64 , 64))
            
    image_temp2 = cv2.imread(row.bedroom_img)
    image2 = cv2.resize(image_temp2, (64 , 64))
            
    image_temp3 = cv2.imread(row.frontal_img)
    image3 = cv2.resize(image_temp3, (64 , 64))
            
    image_temp4 = cv2.imread(row.kitchen_img)
    image4 = cv2.resize(image_temp4, (64 , 64))
              
    inputImages.append(image1)
    inputImages.append(image2)
    inputImages.append(image3)
    inputImages.append(image4)
            
    outputImage[0:64, 0:64] = inputImages[0]
    outputImage[0:64, 64:128] = inputImages[1]
    outputImage[64:128, 64:128] = inputImages[2]
    outputImage[64:128, 0:64] = inputImages[3]
            
    images_output.append(outputImage)
In [28]:
plt.figure(figsize=(8,8))
plt.imshow(images_output[0], interpolation='nearest')
plt.show()
No description has been provided for this image

Check if same outliers are removed from both img and textual tables¶

In [29]:
# All the images match
value = data[img_df.index == data.index]
value.shape
Out[29]:
(308, 5)

Normalize/One Hot Encode Textual Data and Normalize image data¶

In [30]:
# One hot encode zip code & area
# encode_text_dummy(data, 'zipcode')
# encode_text_dummy(data, 'area')

data_formatted = pd.get_dummies(data, columns=['zipcode'])
In [31]:
data_formatted.head()
Out[31]:
Bedrooms Bathrooms area price zipcode_36372 zipcode_60002 zipcode_62025 zipcode_62249 zipcode_81418 zipcode_81524 ... zipcode_93314 zipcode_93446 zipcode_93510 zipcode_94501 zipcode_94531 zipcode_94565 zipcode_94568 zipcode_95220 zipcode_96019 zipcode_98021
0 4 4.0 4053 869500 False False False False False False ... False False False False False False False False False False
1 4 3.0 3343 865200 True False False False False False ... False False False False False False False False False False
2 3 4.0 3923 889000 False False False False False False ... False False False False False False False False False False
6 3 4.0 2544 799000 False False False False False False ... False False False False False False False False False False
10 5 5.0 4829 519200 False False False False False False ... False False False False False False False False False False

5 rows × 36 columns

In [32]:
# Normalize bedrooms & bathrooms & area
encode_numeric_zscore(data_formatted, 'Bedrooms')
encode_numeric_zscore(data_formatted, 'Bathrooms')
encode_numeric_zscore(data_formatted, 'area')
In [33]:
print(data_formatted['price'].dtype)

data_formatted['price'] = data_formatted['price'].astype("float32")

print(data_formatted['price'].dtype)

int64
float32
In [34]:
# data_formatted['price'] = data_formatted['price']
In [35]:
# normalize image data
img_output_df = np.array(images_output).astype('float32')
img_output_df /= 255.0

to_xy and Test/Train Split¶

In [36]:
# use to_xy to split the tabular data into x and y
x, y = to_xy(data_formatted, 'price')
print(x.shape, y.shape)

(308, 35) (308,)
In [37]:
y[:5]
Out[37]:
array([869500., 865200., 889000., 799000., 519200.], dtype=float32)
In [38]:
x_train, x_test, y_train, y_test= train_test_split(x, y, test_size=.20, random_state = 42)
In [39]:
print("Train Sizes :", x_train.shape, y_train.shape)
print("Test Sizes :", x_test.shape, y_test.shape)

Train Sizes : (246, 35) (246,)
Test Sizes : (62, 35) (62,)
In [40]:
x_train_img, x_test_img = train_test_split(img_output_df, test_size=.20,
                                          random_state=42)
In [41]:
print("Train Sizes :", x_train_img.shape)
print("Test Sizes :", x_test_img.shape)

Train Sizes : (246, 128, 128, 3)
Test Sizes : (62, 128, 128, 3)

Pre Model Training Check¶

In [42]:
print(f"x_train shape: ", x_train.shape)
print(f"x_train_img shape: ", x_train_img.shape)
print(f"x_test shape: ", x_test.shape)
print(f"x_test_img shape: ", x_test_img.shape)
print()

print(f"y_train shape: ", y_train.shape)
print(f"y_test shape: ",y_test.shape)

x_train shape:  (246, 35)
x_train_img shape:  (246, 128, 128, 3)
x_test shape:  (62, 35)
x_test_img shape:  (62, 128, 128, 3)

y_train shape:  (246,)
y_test shape:  (62,)
In [43]:
from keras.utils import plot_model
from tensorflow.keras.layers import Conv2D, MaxPooling2D, Flatten
from tensorflow.keras.layers import Input, Dense
from tensorflow.keras.models import Model, Sequential
from tensorflow.keras.layers import concatenate
from keras.optimizers import Adam, SGD

Base Model : Only Textual Input¶

In [44]:
filepath = './dnn/best_weights-base.keras'
checkpointer = ModelCheckpoint(filepath=filepath, verbose=0, save_best_only=True) 

adam_optimizer = Adam(learning_rate=0.0075, beta_1=0.9, beta_2=0.999,epsilon=1e-07)
sgd_optimizer = SGD(learning_rate=0.007, momentum=0.9, nesterov=True)
monitor = EarlyStopping(monitor='val_loss', min_delta=1e-5, patience=5,verbose=1, mode='auto')

model = Sequential()
model.add(Dense(4096, activation='relu'))
model.add(Dense(3000, activation='relu'))
model.add(Dropout(0.3))

model.add(Dense(2048, activation='relu'))
model.add(Dense(1024, activation='relu'))
model.add(Dense(516, activation='relu'))
model.add(Dense(1))

# print(model.summary())
# plot_model(model, show_shapes=True)
model.compile(loss='mean_squared_error', optimizer=adam_optimizer)
model.fit(x_train, y_train, 
          validation_data=(x_test, y_test), 
          callbacks=[monitor, checkpointer],verbose=2,epochs=200)
  
print('Training model 1 finished...Loading the best model')
print()
model.load_weights(filepath) # load weights from best model

pred_base = model.predict(x_test)
score_base = np.sqrt(metrics.mean_squared_error(pred_base,y_test))
print("Score (RMSE): {}".format(score_base))

Epoch 1/200

WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1729645961.509953     101 service.cc:145] XLA service 0x7acb28006fe0 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
I0000 00:00:1729645961.509998     101 service.cc:153]   StreamExecutor device (0): Tesla T4, Compute Capability 7.5
I0000 00:00:1729645961.510003     101 service.cc:153]   StreamExecutor device (1): Tesla T4, Compute Capability 7.5
I0000 00:00:1729645964.844040     101 device_compiler.h:188] Compiled cluster using XLA!  This line is logged at most once for the lifetime of the process.

8/8 - 16s - 2s/step - loss: 205308133376.0000 - val_loss: 71286177792.0000
Epoch 2/200
8/8 - 1s - 153ms/step - loss: 39483125760.0000 - val_loss: 42175311872.0000
Epoch 3/200
8/8 - 1s - 144ms/step - loss: 41181831168.0000 - val_loss: 32654471168.0000
Epoch 4/200
8/8 - 0s - 13ms/step - loss: 31822434304.0000 - val_loss: 44866752512.0000
Epoch 5/200
8/8 - 1s - 148ms/step - loss: 23962263552.0000 - val_loss: 22959218688.0000
Epoch 6/200
8/8 - 4s - 487ms/step - loss: 20260399104.0000 - val_loss: 19696099328.0000
Epoch 7/200
8/8 - 0s - 13ms/step - loss: 20689192960.0000 - val_loss: 31720919040.0000
Epoch 8/200
8/8 - 0s - 12ms/step - loss: 20090097664.0000 - val_loss: 22250913792.0000
Epoch 9/200
8/8 - 0s - 12ms/step - loss: 17019675648.0000 - val_loss: 27193542656.0000
Epoch 10/200
8/8 - 1s - 152ms/step - loss: 13604897792.0000 - val_loss: 19151042560.0000
Epoch 11/200
8/8 - 2s - 259ms/step - loss: 10825491456.0000 - val_loss: 18894815232.0000
Epoch 12/200
8/8 - 0s - 13ms/step - loss: 10533401600.0000 - val_loss: 19779686400.0000
Epoch 13/200
8/8 - 0s - 12ms/step - loss: 14451417088.0000 - val_loss: 20516184064.0000
Epoch 14/200
8/8 - 0s - 12ms/step - loss: 12054653952.0000 - val_loss: 21815433216.0000
Epoch 15/200
8/8 - 0s - 12ms/step - loss: 10899614720.0000 - val_loss: 23821811712.0000
Epoch 16/200
8/8 - 1s - 149ms/step - loss: 13756422144.0000 - val_loss: 17567539200.0000
Epoch 17/200
8/8 - 0s - 12ms/step - loss: 15954426880.0000 - val_loss: 27057399808.0000
Epoch 18/200
8/8 - 0s - 12ms/step - loss: 13367968768.0000 - val_loss: 24133998592.0000
Epoch 19/200
8/8 - 0s - 12ms/step - loss: 14764444672.0000 - val_loss: 19979614208.0000
Epoch 20/200
8/8 - 0s - 11ms/step - loss: 12718134272.0000 - val_loss: 20548304896.0000
Epoch 21/200
8/8 - 0s - 11ms/step - loss: 11548692480.0000 - val_loss: 22658754560.0000
Epoch 21: early stopping
Training model 1 finished...Loading the best model

2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 187ms/step
Score (RMSE): 132542.609375
In [45]:
chart_regression(pred_base.flatten(),y_test)
No description has been provided for this image

Both Input : Model #1¶

In [46]:
filepath = './dnn/best_weights1.keras'
checkpointer = ModelCheckpoint(filepath=filepath, verbose=0, save_best_only=True) 

adam_optimizer = Adam(learning_rate=0.005, beta_1=0.9, beta_2=0.999, epsilon=1e-07)
sgd_optimizer = SGD(learning_rate=0.005, momentum=0.9, nesterov=True)


monitor = EarlyStopping(monitor='val_loss', min_delta=1e-3, patience=5,verbose=1, mode='auto')

# first image input model
visible1 = Input(shape=(128, 128, 3))
conv11 = Conv2D(32, kernel_size=4, activation='relu')(visible1)
pool11 = MaxPooling2D(pool_size=(2, 2))(conv11)
conv12 = Conv2D(64, kernel_size=4, activation='relu')(pool11)
pool12 = MaxPooling2D(pool_size=(2, 2))(conv12)
# conv13 = Conv2D(64, kernel_size=4, activation='relu')(pool12)
# pool13 = MaxPooling2D(pool_size=(2, 2))(conv13)
flat1 = Flatten()(pool12)

# Tabular data input model 
visible2 = Input(shape=(x.shape[1],))
dense1 = Dense(128, activation='relu')(visible2)
dense2 = Dense(64, activation='relu')(dense1)
dense3 = Dense(32, activation='relu')(dense2)

merge = concatenate([flat1, dense3])

# hidden1 = Dense(32, activation='relu')(merge)
# hidden2 = Dense(8, activation='relu')(hidden1)
# output = Dense(1)(hidden2)

output = Dense(1)(merge)

model = Model(inputs=[visible1, visible2], outputs=output)

# print(model.summary())
# plot_model(model, show_shapes=True)
model.compile(loss='mean_squared_error', optimizer=adam_optimizer)
model.fit([x_train_img, x_train], y_train, 
          validation_data=([x_test_img, x_test], y_test), 
          callbacks=[monitor, checkpointer],verbose=2,epochs=100)
  
print('Training model 1 finished...Loading the best model')
print()
model.load_weights(filepath) # load weights from best model

pred = model.predict([x_test_img, x_test])
score = np.sqrt(metrics.mean_squared_error(pred,y_test))
print("Score (RMSE): {}".format(score))

Epoch 1/100
8/8 - 8s - 1s/step - loss: 355371155456.0000 - val_loss: 359970963456.0000
Epoch 2/100
8/8 - 0s - 27ms/step - loss: 316526166016.0000 - val_loss: 258217508864.0000
Epoch 3/100
8/8 - 0s - 26ms/step - loss: 157490135040.0000 - val_loss: 40245022720.0000
Epoch 4/100
8/8 - 0s - 22ms/step - loss: 43660349440.0000 - val_loss: 59467612160.0000
Epoch 5/100
8/8 - 0s - 28ms/step - loss: 35723415552.0000 - val_loss: 35813642240.0000
Epoch 6/100
8/8 - 0s - 27ms/step - loss: 35536146432.0000 - val_loss: 32045844480.0000
Epoch 7/100
8/8 - 0s - 31ms/step - loss: 28900614144.0000 - val_loss: 32354478080.0000
Epoch 8/100
8/8 - 0s - 42ms/step - loss: 28354785280.0000 - val_loss: 29548832768.0000
Epoch 9/100
8/8 - 0s - 25ms/step - loss: 27668094976.0000 - val_loss: 29022355456.0000
Epoch 10/100
8/8 - 0s - 20ms/step - loss: 27189727232.0000 - val_loss: 29022423040.0000
Epoch 11/100
8/8 - 0s - 20ms/step - loss: 26892947456.0000 - val_loss: 29162377216.0000
Epoch 12/100
8/8 - 0s - 25ms/step - loss: 26572111872.0000 - val_loss: 28654704640.0000
Epoch 13/100
8/8 - 0s - 25ms/step - loss: 26343370752.0000 - val_loss: 28191791104.0000
Epoch 14/100
8/8 - 0s - 25ms/step - loss: 25341939712.0000 - val_loss: 27576506368.0000
Epoch 15/100
8/8 - 0s - 25ms/step - loss: 24539777024.0000 - val_loss: 26973394944.0000
Epoch 16/100
8/8 - 0s - 25ms/step - loss: 23434692608.0000 - val_loss: 26428973056.0000
Epoch 17/100
8/8 - 0s - 25ms/step - loss: 22721490944.0000 - val_loss: 25822529536.0000
Epoch 18/100
8/8 - 0s - 25ms/step - loss: 22040862720.0000 - val_loss: 25555359744.0000
Epoch 19/100
8/8 - 0s - 25ms/step - loss: 20982036480.0000 - val_loss: 24436015104.0000
Epoch 20/100
8/8 - 0s - 25ms/step - loss: 20656363520.0000 - val_loss: 23632744448.0000
Epoch 21/100
8/8 - 0s - 25ms/step - loss: 19649619968.0000 - val_loss: 23087493120.0000
Epoch 22/100
8/8 - 0s - 25ms/step - loss: 18815162368.0000 - val_loss: 22236049408.0000
Epoch 23/100
8/8 - 0s - 25ms/step - loss: 18231007232.0000 - val_loss: 21399592960.0000
Epoch 24/100
8/8 - 0s - 25ms/step - loss: 17510416384.0000 - val_loss: 20707239936.0000
Epoch 25/100
8/8 - 0s - 25ms/step - loss: 16709104640.0000 - val_loss: 20381603840.0000
Epoch 26/100
8/8 - 0s - 27ms/step - loss: 16269428736.0000 - val_loss: 19677763584.0000
Epoch 27/100
8/8 - 0s - 25ms/step - loss: 15813370880.0000 - val_loss: 19269513216.0000
Epoch 28/100
8/8 - 0s - 25ms/step - loss: 15320029184.0000 - val_loss: 18756499456.0000
Epoch 29/100
8/8 - 0s - 25ms/step - loss: 15028980736.0000 - val_loss: 18404753408.0000
Epoch 30/100
8/8 - 0s - 25ms/step - loss: 14922741760.0000 - val_loss: 18146000896.0000
Epoch 31/100
8/8 - 0s - 25ms/step - loss: 14087390208.0000 - val_loss: 17947486208.0000
Epoch 32/100
8/8 - 0s - 25ms/step - loss: 14019723264.0000 - val_loss: 17645209600.0000
Epoch 33/100
8/8 - 0s - 25ms/step - loss: 14538913792.0000 - val_loss: 17450928128.0000
Epoch 34/100
8/8 - 0s - 25ms/step - loss: 13422679040.0000 - val_loss: 17424601088.0000
Epoch 35/100
8/8 - 0s - 25ms/step - loss: 13276497920.0000 - val_loss: 17175976960.0000
Epoch 36/100
8/8 - 0s - 25ms/step - loss: 12968394752.0000 - val_loss: 17128804352.0000
Epoch 37/100
8/8 - 0s - 25ms/step - loss: 12886464512.0000 - val_loss: 16861050880.0000
Epoch 38/100
8/8 - 0s - 25ms/step - loss: 12606860288.0000 - val_loss: 16709944320.0000
Epoch 39/100
8/8 - 0s - 25ms/step - loss: 12504014848.0000 - val_loss: 16673518592.0000
Epoch 40/100
8/8 - 0s - 25ms/step - loss: 12283606016.0000 - val_loss: 16484448256.0000
Epoch 41/100
8/8 - 0s - 25ms/step - loss: 12126420992.0000 - val_loss: 16355249152.0000
Epoch 42/100
8/8 - 0s - 25ms/step - loss: 12039969792.0000 - val_loss: 16233763840.0000
Epoch 43/100
8/8 - 0s - 25ms/step - loss: 11898134528.0000 - val_loss: 16059382784.0000
Epoch 44/100
8/8 - 0s - 25ms/step - loss: 11793110016.0000 - val_loss: 15931476992.0000
Epoch 45/100
8/8 - 0s - 20ms/step - loss: 11659071488.0000 - val_loss: 15943547904.0000
Epoch 46/100
8/8 - 0s - 26ms/step - loss: 11519364096.0000 - val_loss: 15823844352.0000
Epoch 47/100
8/8 - 0s - 25ms/step - loss: 11355092992.0000 - val_loss: 15711588352.0000
Epoch 48/100
8/8 - 0s - 25ms/step - loss: 11425711104.0000 - val_loss: 15675561984.0000
Epoch 49/100
8/8 - 0s - 25ms/step - loss: 11325589504.0000 - val_loss: 15646523392.0000
Epoch 50/100
8/8 - 0s - 26ms/step - loss: 11186330624.0000 - val_loss: 15547138048.0000
Epoch 51/100
8/8 - 0s - 20ms/step - loss: 11051229184.0000 - val_loss: 15725873152.0000
Epoch 52/100
8/8 - 0s - 25ms/step - loss: 11124929536.0000 - val_loss: 15469293568.0000
Epoch 53/100
8/8 - 0s - 20ms/step - loss: 10650782720.0000 - val_loss: 15600510976.0000
Epoch 54/100
8/8 - 0s - 25ms/step - loss: 10923679744.0000 - val_loss: 15252531200.0000
Epoch 55/100
8/8 - 0s - 25ms/step - loss: 10619496448.0000 - val_loss: 14981457920.0000
Epoch 56/100
8/8 - 0s - 25ms/step - loss: 10490962944.0000 - val_loss: 14876012544.0000
Epoch 57/100
8/8 - 0s - 25ms/step - loss: 10423516160.0000 - val_loss: 14791948288.0000
Epoch 58/100
8/8 - 0s - 25ms/step - loss: 10307610624.0000 - val_loss: 14748005376.0000
Epoch 59/100
8/8 - 0s - 25ms/step - loss: 10472237056.0000 - val_loss: 14713533440.0000
Epoch 60/100
8/8 - 0s - 20ms/step - loss: 10775097344.0000 - val_loss: 14745447424.0000
Epoch 61/100
8/8 - 0s - 25ms/step - loss: 10119738368.0000 - val_loss: 14534434816.0000
Epoch 62/100
8/8 - 0s - 25ms/step - loss: 10100016128.0000 - val_loss: 14412533760.0000
Epoch 63/100
8/8 - 0s - 20ms/step - loss: 9931453440.0000 - val_loss: 14424089600.0000
Epoch 64/100
8/8 - 0s - 25ms/step - loss: 9831329792.0000 - val_loss: 14192845824.0000
Epoch 65/100
8/8 - 0s - 25ms/step - loss: 9772566528.0000 - val_loss: 14097193984.0000
Epoch 66/100
8/8 - 0s - 25ms/step - loss: 9726860288.0000 - val_loss: 13966750720.0000
Epoch 67/100
8/8 - 0s - 25ms/step - loss: 9681019904.0000 - val_loss: 13944439808.0000
Epoch 68/100
8/8 - 0s - 20ms/step - loss: 9644004352.0000 - val_loss: 13981966336.0000
Epoch 69/100
8/8 - 0s - 19ms/step - loss: 9518442496.0000 - val_loss: 14147939328.0000
Epoch 70/100
8/8 - 0s - 25ms/step - loss: 9457572864.0000 - val_loss: 13819728896.0000
Epoch 71/100
8/8 - 0s - 25ms/step - loss: 9426213888.0000 - val_loss: 13745347584.0000
Epoch 72/100
8/8 - 0s - 20ms/step - loss: 9497756672.0000 - val_loss: 13820730368.0000
Epoch 73/100
8/8 - 0s - 20ms/step - loss: 9376811008.0000 - val_loss: 13838429184.0000
Epoch 74/100
8/8 - 0s - 26ms/step - loss: 9283536896.0000 - val_loss: 13637082112.0000
Epoch 75/100
8/8 - 0s - 25ms/step - loss: 9224225792.0000 - val_loss: 13606193152.0000
Epoch 76/100
8/8 - 0s - 25ms/step - loss: 9190806528.0000 - val_loss: 13496722432.0000
Epoch 77/100
8/8 - 0s - 37ms/step - loss: 9388322816.0000 - val_loss: 13451107328.0000
Epoch 78/100
8/8 - 0s - 21ms/step - loss: 9206947840.0000 - val_loss: 13689434112.0000
Epoch 79/100
8/8 - 0s - 25ms/step - loss: 9123401728.0000 - val_loss: 13294406656.0000
Epoch 80/100
8/8 - 0s - 20ms/step - loss: 9035085824.0000 - val_loss: 13615443968.0000
Epoch 81/100
8/8 - 0s - 25ms/step - loss: 9191953408.0000 - val_loss: 13180433408.0000
Epoch 82/100
8/8 - 0s - 20ms/step - loss: 8914309120.0000 - val_loss: 13584086016.0000
Epoch 83/100
8/8 - 0s - 25ms/step - loss: 8807494656.0000 - val_loss: 13172307968.0000
Epoch 84/100
8/8 - 0s - 20ms/step - loss: 8837336064.0000 - val_loss: 13224507392.0000
Epoch 85/100
8/8 - 0s - 20ms/step - loss: 8716997632.0000 - val_loss: 13194428416.0000
Epoch 86/100
8/8 - 0s - 25ms/step - loss: 8721318912.0000 - val_loss: 13105902592.0000
Epoch 87/100
8/8 - 0s - 25ms/step - loss: 8813201408.0000 - val_loss: 13056398336.0000
Epoch 88/100
8/8 - 0s - 20ms/step - loss: 8745939968.0000 - val_loss: 13194889216.0000
Epoch 89/100
8/8 - 0s - 25ms/step - loss: 8778481664.0000 - val_loss: 12853784576.0000
Epoch 90/100
8/8 - 0s - 20ms/step - loss: 8677250048.0000 - val_loss: 13211665408.0000
Epoch 91/100
8/8 - 0s - 20ms/step - loss: 8737598464.0000 - val_loss: 12879473664.0000
Epoch 92/100
8/8 - 0s - 20ms/step - loss: 8437882368.0000 - val_loss: 13241866240.0000
Epoch 93/100
8/8 - 0s - 20ms/step - loss: 8589302784.0000 - val_loss: 12917735424.0000
Epoch 94/100
8/8 - 0s - 25ms/step - loss: 8457161728.0000 - val_loss: 12696206336.0000
Epoch 95/100
8/8 - 0s - 20ms/step - loss: 8408297472.0000 - val_loss: 12766589952.0000
Epoch 96/100
8/8 - 0s - 25ms/step - loss: 8442716672.0000 - val_loss: 12638266368.0000
Epoch 97/100
8/8 - 0s - 25ms/step - loss: 8370918400.0000 - val_loss: 12414097408.0000
Epoch 98/100
8/8 - 0s - 20ms/step - loss: 8351080960.0000 - val_loss: 12703996928.0000
Epoch 99/100
8/8 - 0s - 20ms/step - loss: 8647768064.0000 - val_loss: 12512877568.0000
Epoch 100/100
8/8 - 0s - 20ms/step - loss: 8431803904.0000 - val_loss: 12491300864.0000
Training model 1 finished...Loading the best model

2/2 ━━━━━━━━━━━━━━━━━━━━ 1s 296ms/step
Score (RMSE): 111418.5703125
In [47]:
chart_regression(pred.flatten(),y_test)
No description has been provided for this image

Both inputs : Model #2¶

In [48]:
filepath = './dnn/best_weights1.keras'
checkpointer = ModelCheckpoint(filepath=filepath, verbose=0, save_best_only=True) 

adam_optimizer = Adam(learning_rate=0.0075, beta_1=0.9, beta_2=0.999, epsilon=1e-07)
sgd_optimizer = SGD(learning_rate=0.01, momentum=0.9, nesterov=True)

# Maybe go between 1e-3 and 1e-2  = .5e-3
monitor = EarlyStopping(monitor='val_loss', min_delta=1e-3, patience=5,verbose=1, mode='auto')

# first image input model
visible1 = Input(shape=(128, 128, 3))
conv11 = Conv2D(32, kernel_size=4, activation='relu')(visible1)
pool11 = MaxPooling2D(pool_size=(2, 2))(conv11)

conv12 = Conv2D(64, kernel_size=4, activation='relu')(pool11)
pool12 = MaxPooling2D(pool_size=(2, 2))(conv12)

conv13 = Conv2D(128, kernel_size=4, activation='relu')(pool12)
pool13 = MaxPooling2D(pool_size=(2, 2))(conv13)

conv14 = Conv2D(256, kernel_size=4, activation='relu')(pool13)
pool14 = MaxPooling2D(pool_size=(2, 2))(conv14)

# pool14 = Dropout(0.5)(pool14)
flat1 = Flatten()(pool14)

# Tabular data input model 
visible2 = Input(shape=(x.shape[1],))
dense1 = Dense(2048, activation='relu')(visible2)
# dense1 = Dropout(0.3)(dense1)
dense2 = Dense(1024, activation='relu')(dense1)
dense3 = Dense(516, activation='relu')(dense2)

merge = concatenate([flat1, dense3])

hidden1 = Dense(1024, activation='relu')(merge)
hidden2 = Dense(512, activation='relu')(hidden1)
hidden2 = Dense(256, activation='relu')(hidden1)
output = Dense(1)(hidden2)

# output = Dense(1)(merge)

model = Model(inputs=[visible1, visible2], outputs=output)

# print(model.summary())
# plot_model(model, show_shapes=True)
model.compile(loss='mean_squared_error', optimizer=adam_optimizer)
model.fit([x_train_img, x_train], y_train, 
          validation_data=([x_test_img, x_test], y_test), 
          callbacks=[monitor, checkpointer],verbose=2,epochs=100)
  
print('Training model 1 finished...Loading the best model')
print()
model.load_weights(filepath) # load weights from best model

pred2 = model.predict([x_test_img, x_test])
score2 = np.sqrt(metrics.mean_squared_error(pred2,y_test))
print("Score (RMSE): {}".format(score2))

Epoch 1/100
8/8 - 15s - 2s/step - loss: 389828804608.0000 - val_loss: 370462785536.0000
Epoch 2/100
8/8 - 1s - 97ms/step - loss: 356003119104.0000 - val_loss: 361044738048.0000
Epoch 3/100
8/8 - 1s - 94ms/step - loss: 247356686336.0000 - val_loss: 74892238848.0000
Epoch 4/100
8/8 - 1s - 93ms/step - loss: 48796364800.0000 - val_loss: 43721646080.0000
Epoch 5/100
8/8 - 1s - 92ms/step - loss: 33067431936.0000 - val_loss: 28155848704.0000
Epoch 6/100
8/8 - 0s - 27ms/step - loss: 27799035904.0000 - val_loss: 32036517888.0000
Epoch 7/100
8/8 - 0s - 27ms/step - loss: 27639185408.0000 - val_loss: 29596291072.0000
Epoch 8/100
8/8 - 0s - 27ms/step - loss: 22858250240.0000 - val_loss: 28584753152.0000
Epoch 9/100
8/8 - 1s - 90ms/step - loss: 20207935488.0000 - val_loss: 18616401920.0000
Epoch 10/100
8/8 - 3s - 374ms/step - loss: 14237435904.0000 - val_loss: 16916883456.0000
Epoch 11/100
8/8 - 1s - 97ms/step - loss: 13455213568.0000 - val_loss: 16301957120.0000
Epoch 12/100
8/8 - 1s - 88ms/step - loss: 14092354560.0000 - val_loss: 11635962880.0000
Epoch 13/100
8/8 - 0s - 28ms/step - loss: 10395865088.0000 - val_loss: 12173178880.0000
Epoch 14/100
8/8 - 0s - 27ms/step - loss: 10911203328.0000 - val_loss: 12494735360.0000
Epoch 15/100
8/8 - 0s - 28ms/step - loss: 9924709376.0000 - val_loss: 12870940672.0000
Epoch 16/100
8/8 - 0s - 28ms/step - loss: 9601597440.0000 - val_loss: 13427461120.0000
Epoch 17/100
8/8 - 1s - 176ms/step - loss: 9325877248.0000 - val_loss: 11597479936.0000
Epoch 18/100
8/8 - 1s - 89ms/step - loss: 8464832000.0000 - val_loss: 11062200320.0000
Epoch 19/100
8/8 - 0s - 29ms/step - loss: 8442086400.0000 - val_loss: 11811104768.0000
Epoch 20/100
8/8 - 0s - 27ms/step - loss: 7618535424.0000 - val_loss: 12601833472.0000
Epoch 21/100
8/8 - 0s - 27ms/step - loss: 8098503168.0000 - val_loss: 11812276224.0000
Epoch 22/100
8/8 - 1s - 90ms/step - loss: 8024230912.0000 - val_loss: 10934122496.0000
Epoch 23/100
8/8 - 0s - 27ms/step - loss: 7319531520.0000 - val_loss: 10959563776.0000
Epoch 24/100
8/8 - 0s - 27ms/step - loss: 8123524096.0000 - val_loss: 12770299904.0000
Epoch 25/100
8/8 - 0s - 27ms/step - loss: 8078976512.0000 - val_loss: 21616267264.0000
Epoch 26/100
8/8 - 0s - 28ms/step - loss: 10584829952.0000 - val_loss: 22934452224.0000
Epoch 27/100
8/8 - 0s - 27ms/step - loss: 10590212096.0000 - val_loss: 14250762240.0000
Epoch 27: early stopping
Training model 1 finished...Loading the best model

2/2 ━━━━━━━━━━━━━━━━━━━━ 1s 418ms/step
Score (RMSE): 104566.34375
In [49]:
chart_regression(pred2.flatten(),y_test)
No description has been provided for this image

Both Inputs : Model # 3¶

In [51]:
filepath = './dnn/best_weights3.keras'
checkpointer = ModelCheckpoint(filepath=filepath, verbose=0, save_best_only=True) 

adam_optimizer = Adam(learning_rate=0.0075, beta_1=0.9, beta_2=0.999, epsilon=1e-07)
sgd_optimizer = SGD(learning_rate=0.01, momentum=0.9, nesterov=True)

# Maybe go between 1e-3 and 1e-2  = .5e-3
monitor = EarlyStopping(monitor='val_loss', min_delta=1e-3, patience=5,verbose=1, mode='auto')

# first image input model
visible1 = Input(shape=(128, 128, 3))
conv11 = Conv2D(32, kernel_size=4, activation='relu')(visible1)
conv11 = Conv2D(32, kernel_size=4, activation='relu')(conv11)
pool11 = MaxPooling2D(pool_size=(2, 2))(conv11)

conv12 = Conv2D(64, kernel_size=4, activation='relu')(pool11)
conv12 = Conv2D(64, kernel_size=4, activation='relu')(conv12)
pool12 = MaxPooling2D(pool_size=(2, 2))(conv12)

conv13 = Conv2D(128, kernel_size=4, activation='relu')(pool12)
conv13 = Conv2D(256, kernel_size=4, activation='relu')(conv12)
pool13 = Dropout(0.5)(conv13) # Why does this work?
pool13 = MaxPooling2D(pool_size=(2, 2))(conv13)

conv14 = Conv2D(512, kernel_size=4, activation='relu')(pool13)
pool14 = MaxPooling2D(pool_size=(2, 2))(conv14)

# pool14 = Dropout(0.5)(pool14)
flat1 = Flatten()(pool14)

# Tabular data input model 
visible2 = Input(shape=(x.shape[1],))
dense1 = Dense(2048, activation='relu')(visible2)
dense1 = Dropout(0.3)(dense1)
dense2 = Dense(1024, activation='relu')(dense1)
dense2 = Dropout(0.5)(dense2)
dense3 = Dense(516, activation='relu')(dense2)

merge = concatenate([flat1, dense3])

hidden1 = Dense(1024, activation='relu')(merge)
hidden2 = Dense(512, activation='relu')(hidden1)
hidden3 = Dense(256, activation='relu')(hidden2)
output = Dense(1)(hidden3)

# output = Dense(1)(merge)

model = Model(inputs=[visible1, visible2], outputs=output)

# print(model.summary())
# plot_model(model, show_shapes=True)
model.compile(loss='mean_squared_error', optimizer=adam_optimizer)
model.fit([x_train_img, x_train], y_train, 
          validation_data=([x_test_img, x_test], y_test), 
          callbacks=[monitor, checkpointer],verbose=2,epochs=100)
  
print('Training model 1 finished...Loading the best model')
print()
model.load_weights(filepath) # load weights from best model

pred3 = model.predict([x_test_img, x_test])
score3 = np.sqrt(metrics.mean_squared_error(pred3,y_test))
print("Score (RMSE): {}".format(score3))

Epoch 1/100
8/8 - 37s - 5s/step - loss: 288773341184.0000 - val_loss: 3572049379328.0000
Epoch 2/100
8/8 - 4s - 561ms/step - loss: 717000146944.0000 - val_loss: 370902532096.0000
Epoch 3/100
8/8 - 7s - 854ms/step - loss: 356601331712.0000 - val_loss: 366871117824.0000
Epoch 4/100
8/8 - 5s - 682ms/step - loss: 250228490240.0000 - val_loss: 80654065664.0000
Epoch 5/100
8/8 - 7s - 861ms/step - loss: 47811641344.0000 - val_loss: 30857465856.0000
Epoch 6/100
8/8 - 5s - 682ms/step - loss: 29340442624.0000 - val_loss: 27620294656.0000
Epoch 7/100
8/8 - 7s - 851ms/step - loss: 21733648384.0000 - val_loss: 22747494400.0000
Epoch 8/100
8/8 - 1s - 116ms/step - loss: 21051240448.0000 - val_loss: 22858186752.0000
Epoch 9/100
8/8 - 4s - 548ms/step - loss: 18788083712.0000 - val_loss: 20866897920.0000
Epoch 10/100
8/8 - 1s - 112ms/step - loss: 16698616832.0000 - val_loss: 23065987072.0000
Epoch 11/100
8/8 - 1s - 105ms/step - loss: 16494222336.0000 - val_loss: 22608250880.0000
Epoch 12/100
8/8 - 4s - 555ms/step - loss: 17219028992.0000 - val_loss: 15397894144.0000
Epoch 13/100
8/8 - 5s - 641ms/step - loss: 12522994688.0000 - val_loss: 15370249216.0000
Epoch 14/100
8/8 - 1s - 112ms/step - loss: 11368299520.0000 - val_loss: 15682735104.0000
Epoch 15/100
8/8 - 7s - 840ms/step - loss: 11442593792.0000 - val_loss: 11902702592.0000
Epoch 16/100
8/8 - 1s - 111ms/step - loss: 11103521792.0000 - val_loss: 11929035776.0000
Epoch 17/100
8/8 - 1s - 104ms/step - loss: 10806233088.0000 - val_loss: 17628491776.0000
Epoch 18/100
8/8 - 1s - 104ms/step - loss: 11756621824.0000 - val_loss: 20629256192.0000
Epoch 19/100
8/8 - 1s - 104ms/step - loss: 12366111744.0000 - val_loss: 14723565568.0000
Epoch 20/100
8/8 - 1s - 106ms/step - loss: 9664669696.0000 - val_loss: 16066801664.0000
Epoch 20: early stopping
Training model 1 finished...Loading the best model

2/2 ━━━━━━━━━━━━━━━━━━━━ 1s 509ms/step
Score (RMSE): 109099.5
In [54]:
chart_regression(pred3.flatten(),y_test)
No description has been provided for this image

Transfer Learning : VGG16 #1¶

In [77]:
# Transfer learning
from keras.applications.vgg16 import VGG16

filepath = './dnn/best_weights-vgg1.keras'
checkpointer = ModelCheckpoint(filepath=filepath, verbose=0, save_best_only=True) 

adam_optimizer = Adam(learning_rate=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-07)
sgd_optimizer = SGD(learning_rate=0.005, momentum=0.9, nesterov=True)

monitor = EarlyStopping(monitor='val_loss', min_delta=1e-3, patience=5, verbose=1, mode='auto')

# Transfer - Learning : VGG 16
vgg = VGG16(weights='imagenet', include_top=False, input_shape=(128, 128, 3)) 
  
# Freeze the weights
for layer in vgg.layers:
    layer.trainable = False

conv_layer = Conv2D(128, kernel_size=3, activation='relu')(vgg.output)  
max_pool = MaxPooling2D(pool_size=(2, 2))(conv_layer)
vgg_out = Flatten()(max_pool)
visible1 = vgg.input

# Tabular data input model 
visible2 = Input(shape=(x.shape[1],))
dense1 = Dense(2048, activation='relu')(visible2)
dense2 = Dense(1024, activation='relu')(dense1)
dense3 = Dense(512, activation='relu')(dense2)

merge = concatenate([vgg_out, dense3])

output = Dense(1)(merge)

model = Model(inputs=[visible1, visible2], outputs=output)

# print(model.summary())
model.compile(loss='mean_squared_error', optimizer=adam_optimizer)
model.fit([x_train_img, x_train], y_train, 
          validation_data=([x_test_img, x_test], y_test), 
          callbacks=[monitor, checkpointer],verbose=2,epochs=100)
  
print('Training finished...Loading the best model')
print()
model.load_weights(filepath) # load weights from best model

pred_vgg1 = model.predict([x_test_img, x_test])
score_vgg1 = np.sqrt(metrics.mean_squared_error(pred_vgg1,y_test))
print("Score (RMSE): {}".format(score_vgg1))

Epoch 1/100
8/8 - 6s - 764ms/step - loss: 357109497856.0000 - val_loss: 370915442688.0000
Epoch 2/100
8/8 - 1s - 113ms/step - loss: 356988944384.0000 - val_loss: 370658705408.0000
Epoch 3/100
8/8 - 1s - 105ms/step - loss: 356540809216.0000 - val_loss: 369840062464.0000
Epoch 4/100
8/8 - 1s - 105ms/step - loss: 355244867584.0000 - val_loss: 367667609600.0000
Epoch 5/100
8/8 - 1s - 109ms/step - loss: 352157007872.0000 - val_loss: 362729996288.0000
Epoch 6/100
8/8 - 1s - 106ms/step - loss: 345199706112.0000 - val_loss: 352804208640.0000
Epoch 7/100
8/8 - 2s - 263ms/step - loss: 331964973056.0000 - val_loss: 334684979200.0000
Epoch 8/100
8/8 - 1s - 105ms/step - loss: 309405581312.0000 - val_loss: 304690692096.0000
Epoch 9/100
8/8 - 1s - 106ms/step - loss: 272897736704.0000 - val_loss: 259861118976.0000
Epoch 10/100
8/8 - 1s - 105ms/step - loss: 221373038592.0000 - val_loss: 200332050432.0000
Epoch 11/100
8/8 - 2s - 299ms/step - loss: 155859075072.0000 - val_loss: 135237328896.0000
Epoch 12/100
8/8 - 1s - 115ms/step - loss: 92154290176.0000 - val_loss: 83900424192.0000
Epoch 13/100
8/8 - 1s - 104ms/step - loss: 51194753024.0000 - val_loss: 67028721664.0000
Epoch 14/100
8/8 - 0s - 55ms/step - loss: 42852528128.0000 - val_loss: 68638695424.0000
Epoch 15/100
8/8 - 1s - 107ms/step - loss: 41574993920.0000 - val_loss: 62080843776.0000
Epoch 16/100
8/8 - 2s - 197ms/step - loss: 36203331584.0000 - val_loss: 54432636928.0000
Epoch 17/100
8/8 - 1s - 106ms/step - loss: 33155434496.0000 - val_loss: 49799630848.0000
Epoch 18/100
8/8 - 1s - 104ms/step - loss: 30410362880.0000 - val_loss: 46167195648.0000
Epoch 19/100
8/8 - 1s - 107ms/step - loss: 28028039168.0000 - val_loss: 42810753024.0000
Epoch 20/100
8/8 - 1s - 106ms/step - loss: 25905510400.0000 - val_loss: 39954575360.0000
Epoch 21/100
8/8 - 2s - 291ms/step - loss: 24549427200.0000 - val_loss: 37577650176.0000
Epoch 22/100
8/8 - 1s - 113ms/step - loss: 22828931072.0000 - val_loss: 35228577792.0000
Epoch 23/100
8/8 - 1s - 108ms/step - loss: 21526882304.0000 - val_loss: 33358602240.0000
Epoch 24/100
8/8 - 1s - 104ms/step - loss: 20345645056.0000 - val_loss: 31762339840.0000
Epoch 25/100
8/8 - 2s - 248ms/step - loss: 19263127552.0000 - val_loss: 30256883712.0000
Epoch 26/100
8/8 - 1s - 106ms/step - loss: 18374680576.0000 - val_loss: 28959184896.0000
Epoch 27/100
8/8 - 1s - 105ms/step - loss: 17565769728.0000 - val_loss: 27787622400.0000
Epoch 28/100
8/8 - 1s - 107ms/step - loss: 16856208384.0000 - val_loss: 26816325632.0000
Epoch 29/100
8/8 - 1s - 105ms/step - loss: 16157313024.0000 - val_loss: 26037690368.0000
Epoch 30/100
8/8 - 2s - 255ms/step - loss: 15721502720.0000 - val_loss: 25220882432.0000
Epoch 31/100
8/8 - 1s - 106ms/step - loss: 15138616320.0000 - val_loss: 24380461056.0000
Epoch 32/100
8/8 - 1s - 109ms/step - loss: 14616924160.0000 - val_loss: 23868805120.0000
Epoch 33/100
8/8 - 1s - 105ms/step - loss: 14378519552.0000 - val_loss: 23376881664.0000
Epoch 34/100
8/8 - 1s - 106ms/step - loss: 13934904320.0000 - val_loss: 22926737408.0000
Epoch 35/100
8/8 - 2s - 198ms/step - loss: 13607005184.0000 - val_loss: 22553686016.0000
Epoch 36/100
8/8 - 1s - 106ms/step - loss: 13331350528.0000 - val_loss: 22229297152.0000
Epoch 37/100
8/8 - 1s - 104ms/step - loss: 13104901120.0000 - val_loss: 21864189952.0000
Epoch 38/100
8/8 - 1s - 103ms/step - loss: 12864657408.0000 - val_loss: 21456535552.0000
Epoch 39/100
8/8 - 1s - 103ms/step - loss: 12636702720.0000 - val_loss: 21239238656.0000
Epoch 40/100
8/8 - 1s - 104ms/step - loss: 12525586432.0000 - val_loss: 21172148224.0000
Epoch 41/100
8/8 - 2s - 205ms/step - loss: 12347839488.0000 - val_loss: 20952756224.0000
Epoch 42/100
8/8 - 1s - 109ms/step - loss: 12157690880.0000 - val_loss: 20701761536.0000
Epoch 43/100
8/8 - 1s - 105ms/step - loss: 12009214976.0000 - val_loss: 20459124736.0000
Epoch 44/100
8/8 - 1s - 105ms/step - loss: 11943318528.0000 - val_loss: 20259586048.0000
Epoch 45/100
8/8 - 2s - 229ms/step - loss: 11836160000.0000 - val_loss: 20171241472.0000
Epoch 46/100
8/8 - 0s - 55ms/step - loss: 11675684864.0000 - val_loss: 20179892224.0000
Epoch 47/100
8/8 - 1s - 106ms/step - loss: 11696166912.0000 - val_loss: 20047366144.0000
Epoch 48/100
8/8 - 1s - 107ms/step - loss: 11498198016.0000 - val_loss: 19999086592.0000
Epoch 49/100
8/8 - 1s - 107ms/step - loss: 11464457216.0000 - val_loss: 19722694656.0000
Epoch 50/100
8/8 - 1s - 105ms/step - loss: 11362777088.0000 - val_loss: 19586461696.0000
Epoch 51/100
8/8 - 2s - 257ms/step - loss: 11306577920.0000 - val_loss: 19510810624.0000
Epoch 52/100
8/8 - 1s - 107ms/step - loss: 11324113920.0000 - val_loss: 19367768064.0000
Epoch 53/100
8/8 - 1s - 105ms/step - loss: 11210382336.0000 - val_loss: 19319060480.0000
Epoch 54/100
8/8 - 1s - 104ms/step - loss: 11159977984.0000 - val_loss: 19204632576.0000
Epoch 55/100
8/8 - 1s - 106ms/step - loss: 11105344512.0000 - val_loss: 19092064256.0000
Epoch 56/100
8/8 - 2s - 256ms/step - loss: 11054464000.0000 - val_loss: 19090503680.0000
Epoch 57/100
8/8 - 0s - 55ms/step - loss: 10996736000.0000 - val_loss: 19174834176.0000
Epoch 58/100
8/8 - 1s - 106ms/step - loss: 10940626944.0000 - val_loss: 19090456576.0000
Epoch 59/100
8/8 - 1s - 104ms/step - loss: 10915001344.0000 - val_loss: 18939678720.0000
Epoch 60/100
8/8 - 1s - 108ms/step - loss: 10868730880.0000 - val_loss: 18882578432.0000
Epoch 61/100
8/8 - 0s - 57ms/step - loss: 10833651712.0000 - val_loss: 18958907392.0000
Epoch 62/100
8/8 - 0s - 54ms/step - loss: 10799295488.0000 - val_loss: 18885236736.0000
Epoch 63/100
8/8 - 1s - 177ms/step - loss: 10750878720.0000 - val_loss: 18799675392.0000
Epoch 64/100
8/8 - 1s - 106ms/step - loss: 10728214528.0000 - val_loss: 18739951616.0000
Epoch 65/100
8/8 - 1s - 104ms/step - loss: 10744617984.0000 - val_loss: 18691692544.0000
Epoch 66/100
8/8 - 0s - 55ms/step - loss: 10772488192.0000 - val_loss: 18951618560.0000
Epoch 67/100
8/8 - 1s - 106ms/step - loss: 10715024384.0000 - val_loss: 18685167616.0000
Epoch 68/100
8/8 - 0s - 55ms/step - loss: 10676811776.0000 - val_loss: 18928121856.0000
Epoch 69/100
8/8 - 0s - 55ms/step - loss: 10643721216.0000 - val_loss: 18756646912.0000
Epoch 70/100
8/8 - 0s - 55ms/step - loss: 10790673408.0000 - val_loss: 18825242624.0000
Epoch 71/100
8/8 - 2s - 205ms/step - loss: 10744395776.0000 - val_loss: 18521911296.0000
Epoch 72/100
8/8 - 0s - 54ms/step - loss: 10577290240.0000 - val_loss: 18604926976.0000
Epoch 73/100
8/8 - 0s - 55ms/step - loss: 10540840960.0000 - val_loss: 18851338240.0000
Epoch 74/100
8/8 - 0s - 55ms/step - loss: 10556408832.0000 - val_loss: 18657253376.0000
Epoch 75/100
8/8 - 0s - 55ms/step - loss: 10484264960.0000 - val_loss: 18559064064.0000
Epoch 76/100
8/8 - 1s - 106ms/step - loss: 10564759552.0000 - val_loss: 18404345856.0000
Epoch 77/100
8/8 - 1s - 111ms/step - loss: 10453718016.0000 - val_loss: 18335598592.0000
Epoch 78/100
8/8 - 2s - 193ms/step - loss: 10518384640.0000 - val_loss: 18327425024.0000
Epoch 79/100
8/8 - 1s - 106ms/step - loss: 10465306624.0000 - val_loss: 18318348288.0000
Epoch 80/100
8/8 - 1s - 106ms/step - loss: 10422926336.0000 - val_loss: 18144315392.0000
Epoch 81/100
8/8 - 0s - 54ms/step - loss: 10387391488.0000 - val_loss: 18175598592.0000
Epoch 82/100
8/8 - 0s - 55ms/step - loss: 10419236864.0000 - val_loss: 18348818432.0000
Epoch 83/100
8/8 - 0s - 55ms/step - loss: 10356454400.0000 - val_loss: 18190667776.0000
Epoch 84/100
8/8 - 0s - 55ms/step - loss: 10399023104.0000 - val_loss: 18207676416.0000
Epoch 85/100
8/8 - 0s - 55ms/step - loss: 10510418944.0000 - val_loss: 18445844480.0000
Epoch 85: early stopping
Training finished...Loading the best model

2/2 ━━━━━━━━━━━━━━━━━━━━ 1s 468ms/step
Score (RMSE): 134700.84375
In [78]:
chart_regression(pred_vgg1.flatten(),y_test)
No description has been provided for this image

Transfer Learning : VGG16 #2 (trainable = true)¶

In [79]:
# Transfer learning
from keras.applications.vgg16 import VGG16

filepath = './dnn/best_weights-vgg2.keras'
checkpointer = ModelCheckpoint(filepath=filepath, verbose=0, save_best_only=True) 

adam_optimizer = Adam(learning_rate=0.005, beta_1=0.9, beta_2=0.999, epsilon=1e-07)
sgd_optimizer = SGD(learning_rate=0.005, momentum=0.9, nesterov=True)

monitor = EarlyStopping(monitor='val_loss', min_delta=1e-3, patience=5, verbose=1, mode='auto')

# Transfer - Learning : VGG 16
vgg = VGG16(weights='imagenet', include_top=False, input_shape=(128, 128, 3)) 
  
# Freeze the weights
for layer in vgg.layers:
    layer.trainable = True

conv_layer = Conv2D(128, kernel_size=3, activation='relu')(vgg.output)  
max_pool = MaxPooling2D(pool_size=(2, 2))(conv_layer)
vgg_out = Flatten()(max_pool)
visible1 = vgg.input

# Tabular data input model 
visible2 = Input(shape=(x.shape[1],))
dense1 = Dense(2048, activation='relu')(visible2)
dense2 = Dense(1024, activation='relu')(dense1)
dense3 = Dense(512, activation='relu')(dense2)

merge = concatenate([vgg_out, dense3])

# hidden1 = Dense(512, activation='relu')(merge)
# hidden2 = Dense(256, activation='relu')(hidden1)
# output = Dense(1)(hidden2)

output = Dense(1)(merge)

model = Model(inputs=[visible1, visible2], outputs=output)

# print(model.summary())
model.compile(loss='mean_squared_error', optimizer=adam_optimizer)
model.fit([x_train_img, x_train], y_train, 
          validation_data=([x_test_img, x_test], y_test), 
          callbacks=[monitor, checkpointer],verbose=2,epochs=100)
  
print('Training finished...Loading the best model')
print()
# model.load_weights(filepath) # load weights from best model

pred_vgg2 = model.predict([x_test_img, x_test])
score_vgg2 = np.sqrt(metrics.mean_squared_error(pred_vgg2,y_test))
print("Score (RMSE): {}".format(score_vgg2))

Epoch 1/100
8/8 - 11s - 1s/step - loss: 98477956510253056.0000 - val_loss: 303063465984.0000
Epoch 2/100
8/8 - 1s - 139ms/step - loss: 2107533688832.0000 - val_loss: 369756176384.0000
Epoch 3/100
8/8 - 1s - 139ms/step - loss: 356539531264.0000 - val_loss: 367158263808.0000
Epoch 4/100
8/8 - 2s - 257ms/step - loss: 269744537600.0000 - val_loss: 240790159360.0000
Epoch 5/100
8/8 - 2s - 239ms/step - loss: 100941979648.0000 - val_loss: 24683929600.0000
Epoch 6/100
8/8 - 1s - 139ms/step - loss: 53171990528.0000 - val_loss: 41044328448.0000
Epoch 7/100
8/8 - 1s - 139ms/step - loss: 36210229248.0000 - val_loss: 41772793856.0000
Epoch 8/100
8/8 - 1s - 139ms/step - loss: 29340475392.0000 - val_loss: 27252961280.0000
Epoch 9/100
8/8 - 2s - 242ms/step - loss: 30167261184.0000 - val_loss: 24614602752.0000
Epoch 10/100
8/8 - 1s - 142ms/step - loss: 36199636992.0000 - val_loss: 26172733440.0000
Epoch 11/100
8/8 - 1s - 140ms/step - loss: 23779014656.0000 - val_loss: 27124082688.0000
Epoch 12/100
8/8 - 3s - 363ms/step - loss: 24989556736.0000 - val_loss: 24239321088.0000
Epoch 13/100
8/8 - 2s - 250ms/step - loss: 24240252928.0000 - val_loss: 24226115584.0000
Epoch 14/100
8/8 - 1s - 140ms/step - loss: 23725441024.0000 - val_loss: 28107677696.0000
Epoch 15/100
8/8 - 1s - 141ms/step - loss: 26673543168.0000 - val_loss: 28147427328.0000
Epoch 16/100
8/8 - 1s - 141ms/step - loss: 25089722368.0000 - val_loss: 28518916096.0000
Epoch 17/100
8/8 - 1s - 140ms/step - loss: 24446169088.0000 - val_loss: 26050238464.0000
Epoch 18/100
8/8 - 2s - 244ms/step - loss: 20350199808.0000 - val_loss: 20663142400.0000
Epoch 19/100
8/8 - 1s - 141ms/step - loss: 16470412288.0000 - val_loss: 22159106048.0000
Epoch 20/100
8/8 - 2s - 248ms/step - loss: 15757091840.0000 - val_loss: 19971659776.0000
Epoch 21/100
8/8 - 3s - 387ms/step - loss: 16336232448.0000 - val_loss: 17648381952.0000
Epoch 22/100
8/8 - 2s - 251ms/step - loss: 14437679104.0000 - val_loss: 16961068032.0000
Epoch 23/100
8/8 - 1s - 141ms/step - loss: 13898177536.0000 - val_loss: 21335939072.0000
Epoch 24/100
8/8 - 2s - 247ms/step - loss: 15474721792.0000 - val_loss: 14793943040.0000
Epoch 25/100
8/8 - 2s - 244ms/step - loss: 11502510080.0000 - val_loss: 14733190144.0000
Epoch 26/100
8/8 - 2s - 241ms/step - loss: 10533182464.0000 - val_loss: 14276973568.0000
Epoch 27/100
8/8 - 3s - 323ms/step - loss: 10446988288.0000 - val_loss: 13208377344.0000
Epoch 28/100
8/8 - 2s - 244ms/step - loss: 10061524992.0000 - val_loss: 12703187968.0000
Epoch 29/100
8/8 - 1s - 140ms/step - loss: 9188671488.0000 - val_loss: 13632504832.0000
Epoch 30/100
8/8 - 2s - 273ms/step - loss: 9257372672.0000 - val_loss: 12013086720.0000
Epoch 31/100
8/8 - 1s - 142ms/step - loss: 9282802688.0000 - val_loss: 12347857920.0000
Epoch 32/100
8/8 - 1s - 140ms/step - loss: 8902043648.0000 - val_loss: 12226070528.0000
Epoch 33/100
8/8 - 1s - 140ms/step - loss: 9030254592.0000 - val_loss: 13165268992.0000
Epoch 34/100
8/8 - 2s - 242ms/step - loss: 9161007104.0000 - val_loss: 11707984896.0000
Epoch 35/100
8/8 - 2s - 243ms/step - loss: 8301892096.0000 - val_loss: 11574075392.0000
Epoch 36/100
8/8 - 1s - 139ms/step - loss: 7976644096.0000 - val_loss: 12180003840.0000
Epoch 37/100
8/8 - 1s - 140ms/step - loss: 8459735552.0000 - val_loss: 13380633600.0000
Epoch 38/100
8/8 - 3s - 354ms/step - loss: 8173901824.0000 - val_loss: 11080372224.0000
Epoch 39/100
8/8 - 1s - 140ms/step - loss: 8150966272.0000 - val_loss: 11342439424.0000
Epoch 40/100
8/8 - 2s - 243ms/step - loss: 8596393984.0000 - val_loss: 10848686080.0000
Epoch 41/100
8/8 - 1s - 139ms/step - loss: 8398706688.0000 - val_loss: 11815157760.0000
Epoch 42/100
8/8 - 1s - 139ms/step - loss: 7826038272.0000 - val_loss: 11413584896.0000
Epoch 43/100
8/8 - 1s - 139ms/step - loss: 7517352448.0000 - val_loss: 12484359168.0000
Epoch 44/100
8/8 - 1s - 139ms/step - loss: 7593542656.0000 - val_loss: 11686118400.0000
Epoch 45/100
8/8 - 1s - 139ms/step - loss: 7215876608.0000 - val_loss: 12243982336.0000
Epoch 45: early stopping
Training finished...Loading the best model

2/2 ━━━━━━━━━━━━━━━━━━━━ 1s 467ms/step
Score (RMSE): 110652.5234375
In [80]:
chart_regression(pred_vgg2.flatten(),y_test)
No description has been provided for this image

Transfer Learning : ResNet50¶

In [87]:
from keras.applications import ResNet50
from keras.layers import Input, Dense, Flatten, Conv2D, Dropout, concatenate, MaxPooling2D
from keras.models import Model
from keras.optimizers import Adam, SGD
from keras.callbacks import ModelCheckpoint, EarlyStopping
import numpy as np
from sklearn import metrics

filepath = './dnn/best_weights-resNet1.keras'
checkpointer = ModelCheckpoint(filepath=filepath, verbose=0, save_best_only=True) 

adam_optimizer = Adam(learning_rate=0.009, beta_1=0.9, beta_2=0.999, epsilon=1e-07)
sgd_optimizer = SGD(learning_rate=0.005, momentum=0.9, nesterov=True)

monitor = EarlyStopping(monitor='val_loss', min_delta=1e-3, patience=25, verbose=1, mode='auto')

resNet = ResNet50(weights='imagenet', include_top=False, input_shape=(128, 128, 3)) 

# Freeze the ResNet layers
for layer in resNet.layers:
    layer.trainable = False
    
conv_layer = Conv2D(128, kernel_size=4, activation='relu')(resNet.output)  
# max_pool = MaxPooling2D(pool_size=(2,2))(conv_layer)
# res_out = Flatten()(max_pool)

flatten_res = Flatten()(conv_layer) 

# Tabular data input (dense layers)
visible2 = Input(shape=(x.shape[1],))
dense1 = Dense(512, activation='relu')(visible2) 
dense2 = Dense(1024, activation='relu')(dense1)
# dense2 = Dropout(0.5)(dense2)
dense3 = Dense(1024, activation='relu')(dense2)

merge = concatenate([flatten_res, dense3])


hidden1 = Dense(2048, activation='relu')(merge)
# hidden1 = Dropout(0.5)(hidden1)  
hidden2 = Dense(1024, activation='relu')(hidden1)
hidden2 = Dense(1024, activation='relu')(hidden2)
# hidden2 = Dropout(0.5)(hidden2)  
hidden3 = Dense(512, activation='relu')(hidden2)

output = Dense(1)(hidden3)
# output = Dense(1)(merge) 

model = Model(inputs=[resNet.input, visible2], outputs=output)
model.compile(optimizer=adam_optimizer, loss='mean_squared_error')
model.fit([x_train_img, x_train], y_train, 
          validation_data=([x_test_img, x_test], y_test), 
          callbacks=[monitor, checkpointer], verbose=2, epochs=100)

print('Training finished...Loading the best model')
model.load_weights(filepath) 

pred_resNet = model.predict([x_test_img, x_test])
score_resNet = np.sqrt(metrics.mean_squared_error(pred_resNet, y_test))
print("Score (RMSE): {}".format(score_resNet))

Epoch 1/100
8/8 - 20s - 3s/step - loss: 343688642560.0000 - val_loss: 368869212160.0000
Epoch 2/100
8/8 - 2s - 208ms/step - loss: 355331997696.0000 - val_loss: 357225103360.0000
Epoch 3/100
8/8 - 2s - 196ms/step - loss: 217283854336.0000 - val_loss: 35435380736.0000
Epoch 4/100
8/8 - 4s - 461ms/step - loss: 81378050048.0000 - val_loss: 34845188096.0000
Epoch 5/100
8/8 - 0s - 60ms/step - loss: 39354314752.0000 - val_loss: 39310811136.0000
Epoch 6/100
8/8 - 2s - 216ms/step - loss: 24057284608.0000 - val_loss: 26119258112.0000
Epoch 7/100
8/8 - 0s - 43ms/step - loss: 18042099712.0000 - val_loss: 27164329984.0000
Epoch 8/100
8/8 - 2s - 242ms/step - loss: 16030982144.0000 - val_loss: 22946465792.0000
Epoch 9/100
8/8 - 2s - 194ms/step - loss: 14523408384.0000 - val_loss: 20112242688.0000
Epoch 10/100
8/8 - 2s - 197ms/step - loss: 11718484992.0000 - val_loss: 17581297664.0000
Epoch 11/100
8/8 - 3s - 369ms/step - loss: 11536602112.0000 - val_loss: 16853346304.0000
Epoch 12/100
8/8 - 2s - 205ms/step - loss: 11934524416.0000 - val_loss: 16775670784.0000
Epoch 13/100
8/8 - 3s - 351ms/step - loss: 10680504320.0000 - val_loss: 16169242624.0000
Epoch 14/100
8/8 - 0s - 57ms/step - loss: 11737072640.0000 - val_loss: 18993100800.0000
Epoch 15/100
8/8 - 0s - 43ms/step - loss: 12300834816.0000 - val_loss: 20013543424.0000
Epoch 16/100
8/8 - 0s - 43ms/step - loss: 12530514944.0000 - val_loss: 19980361728.0000
Epoch 17/100
8/8 - 0s - 42ms/step - loss: 10984398848.0000 - val_loss: 18687203328.0000
Epoch 18/100
8/8 - 0s - 42ms/step - loss: 10752083968.0000 - val_loss: 18892447744.0000
Epoch 19/100
8/8 - 2s - 199ms/step - loss: 9637939200.0000 - val_loss: 12993325056.0000
Epoch 20/100
8/8 - 0s - 43ms/step - loss: 8086447616.0000 - val_loss: 13412659200.0000
Epoch 21/100
8/8 - 2s - 198ms/step - loss: 8887006208.0000 - val_loss: 12339708928.0000
Epoch 22/100
8/8 - 2s - 192ms/step - loss: 7603577856.0000 - val_loss: 12130540544.0000
Epoch 23/100
8/8 - 0s - 43ms/step - loss: 7191845888.0000 - val_loss: 12283540480.0000
Epoch 24/100
8/8 - 2s - 197ms/step - loss: 7630202880.0000 - val_loss: 11082414080.0000
Epoch 25/100
8/8 - 0s - 43ms/step - loss: 7266596864.0000 - val_loss: 11426635776.0000
Epoch 26/100
8/8 - 0s - 44ms/step - loss: 6850041856.0000 - val_loss: 11277194240.0000
Epoch 27/100
8/8 - 2s - 226ms/step - loss: 7229795840.0000 - val_loss: 11029869568.0000
Epoch 28/100
8/8 - 2s - 199ms/step - loss: 6804054016.0000 - val_loss: 10288318464.0000
Epoch 29/100
8/8 - 0s - 43ms/step - loss: 6173859328.0000 - val_loss: 11615148032.0000
Epoch 30/100
8/8 - 0s - 43ms/step - loss: 6797023232.0000 - val_loss: 12913753088.0000
Epoch 31/100
8/8 - 0s - 43ms/step - loss: 7227819520.0000 - val_loss: 10830779392.0000
Epoch 32/100
8/8 - 0s - 43ms/step - loss: 8002421760.0000 - val_loss: 10657727488.0000
Epoch 33/100
8/8 - 0s - 43ms/step - loss: 6749570560.0000 - val_loss: 16542325760.0000
Epoch 34/100
8/8 - 0s - 43ms/step - loss: 9672331264.0000 - val_loss: 17578686464.0000
Epoch 35/100
8/8 - 2s - 279ms/step - loss: 7330137088.0000 - val_loss: 10097063936.0000
Epoch 36/100
8/8 - 2s - 208ms/step - loss: 6959677440.0000 - val_loss: 9678855168.0000
Epoch 37/100
8/8 - 0s - 44ms/step - loss: 7266845184.0000 - val_loss: 12130420736.0000
Epoch 38/100
8/8 - 3s - 350ms/step - loss: 5950133248.0000 - val_loss: 9031878656.0000
Epoch 39/100
8/8 - 0s - 56ms/step - loss: 4810904576.0000 - val_loss: 11541922816.0000
Epoch 40/100
8/8 - 0s - 45ms/step - loss: 5472159744.0000 - val_loss: 10176888832.0000
Epoch 41/100
8/8 - 0s - 43ms/step - loss: 5512726016.0000 - val_loss: 10960668672.0000
Epoch 42/100
8/8 - 0s - 43ms/step - loss: 6202236928.0000 - val_loss: 12790712320.0000
Epoch 43/100
8/8 - 0s - 43ms/step - loss: 6791001600.0000 - val_loss: 12081680384.0000
Epoch 44/100
8/8 - 0s - 43ms/step - loss: 4952603648.0000 - val_loss: 13286253568.0000
Epoch 45/100
8/8 - 2s - 233ms/step - loss: 5825606144.0000 - val_loss: 8345907200.0000
Epoch 46/100
8/8 - 0s - 44ms/step - loss: 4956800000.0000 - val_loss: 9456014336.0000
Epoch 47/100
8/8 - 0s - 43ms/step - loss: 3720694016.0000 - val_loss: 10123528192.0000
Epoch 48/100
8/8 - 0s - 43ms/step - loss: 3572302336.0000 - val_loss: 9395789824.0000
Epoch 49/100
8/8 - 2s - 200ms/step - loss: 5560272896.0000 - val_loss: 7923381760.0000
Epoch 50/100
8/8 - 0s - 43ms/step - loss: 4261129216.0000 - val_loss: 10889527296.0000
Epoch 51/100
8/8 - 0s - 43ms/step - loss: 5020419072.0000 - val_loss: 11168991232.0000
Epoch 52/100
8/8 - 0s - 43ms/step - loss: 7489695232.0000 - val_loss: 9531190272.0000
Epoch 53/100
8/8 - 0s - 43ms/step - loss: 4580328448.0000 - val_loss: 9275280384.0000
Epoch 54/100
8/8 - 0s - 43ms/step - loss: 3675055616.0000 - val_loss: 8823876608.0000
Epoch 55/100
8/8 - 2s - 222ms/step - loss: 3029416960.0000 - val_loss: 7804430848.0000
Epoch 56/100
8/8 - 0s - 44ms/step - loss: 3477314304.0000 - val_loss: 8397595136.0000
Epoch 57/100
8/8 - 0s - 43ms/step - loss: 3851264512.0000 - val_loss: 10806586368.0000
Epoch 58/100
8/8 - 0s - 43ms/step - loss: 4168550656.0000 - val_loss: 10045679616.0000
Epoch 59/100
8/8 - 0s - 43ms/step - loss: 3604981504.0000 - val_loss: 8787392512.0000
Epoch 60/100
8/8 - 0s - 43ms/step - loss: 3183405568.0000 - val_loss: 11372168192.0000
Epoch 61/100
8/8 - 0s - 43ms/step - loss: 4860128768.0000 - val_loss: 10329928704.0000
Epoch 62/100
8/8 - 0s - 43ms/step - loss: 3874722304.0000 - val_loss: 9601417216.0000
Epoch 63/100
8/8 - 0s - 43ms/step - loss: 4176076032.0000 - val_loss: 8450109440.0000
Epoch 64/100
8/8 - 0s - 43ms/step - loss: 3569046016.0000 - val_loss: 8636995584.0000
Epoch 65/100
8/8 - 0s - 43ms/step - loss: 2799125760.0000 - val_loss: 8914883584.0000
Epoch 66/100
8/8 - 0s - 43ms/step - loss: 2769716224.0000 - val_loss: 8153819136.0000
Epoch 67/100
8/8 - 0s - 43ms/step - loss: 2366477568.0000 - val_loss: 7894991360.0000
Epoch 68/100
8/8 - 0s - 43ms/step - loss: 1827070208.0000 - val_loss: 9157101568.0000
Epoch 69/100
8/8 - 0s - 44ms/step - loss: 2456001280.0000 - val_loss: 7882080768.0000
Epoch 70/100
8/8 - 0s - 43ms/step - loss: 1920451200.0000 - val_loss: 8750027776.0000
Epoch 71/100
8/8 - 0s - 43ms/step - loss: 2360403712.0000 - val_loss: 9532099584.0000
Epoch 72/100
8/8 - 0s - 43ms/step - loss: 1908828416.0000 - val_loss: 9047440384.0000
Epoch 73/100
8/8 - 0s - 44ms/step - loss: 1808188928.0000 - val_loss: 8158190592.0000
Epoch 74/100
8/8 - 0s - 44ms/step - loss: 1572947328.0000 - val_loss: 8420847104.0000
Epoch 75/100
8/8 - 0s - 43ms/step - loss: 1731808000.0000 - val_loss: 9543692288.0000
Epoch 76/100
8/8 - 0s - 43ms/step - loss: 2079366144.0000 - val_loss: 8558274560.0000
Epoch 77/100
8/8 - 0s - 43ms/step - loss: 1567895296.0000 - val_loss: 9269611520.0000
Epoch 78/100
8/8 - 0s - 43ms/step - loss: 1627222784.0000 - val_loss: 8911315968.0000
Epoch 79/100
8/8 - 0s - 43ms/step - loss: 1182259712.0000 - val_loss: 9657309184.0000
Epoch 80/100
8/8 - 0s - 43ms/step - loss: 1372559872.0000 - val_loss: 9690188800.0000
Epoch 80: early stopping
Training finished...Loading the best model
2/2 ━━━━━━━━━━━━━━━━━━━━ 9s 5s/step
Score (RMSE): 88342.6953125
In [64]:
chart_regression(pred_resNet.flatten(),y_test)
No description has been provided for this image

Transfer Learning : ResNet50 #2¶

In [83]:
# Transfer learning
from keras.applications import ResNet50

filepath = './dnn/best_weights-resNet2.keras'
checkpointer = ModelCheckpoint(filepath=filepath, verbose=0, save_best_only=True)

for i in range(0,5): 
    print("======================================= " , i)
    adam_optimizer = Adam(learning_rate=0.0075, beta_1=0.9, beta_2=0.999, epsilon=1e-07)
    sgd_optimizer = SGD(learning_rate=1e-4 , momentum=0.9, nesterov=True)
    monitor = EarlyStopping(monitor='val_loss', min_delta=1e-3, patience=25, verbose=1, mode='auto')
    
    resNet = ResNet50(weights='imagenet', include_top=False, input_shape=(128, 128, 3)) 

    for layer in resNet.layers:
        layer.trainable = False

    conv_layer = Conv2D(128, kernel_size=4, activation='relu')(resNet.output)  
    flatten_res = Flatten()(conv_layer)  


    visible1 = resNet.input

    # Tabular data input model
    visible2 = Input(shape=(x.shape[1],))
    dense1 = Dense(512, activation='relu')(visible2)
    dense2 = Dense(1024, activation='relu')(dense1)
    dense3 = Dense(2048, activation='relu')(dense2)
    dense4 = Dense(512, activation='relu')(dense3)

    merge = concatenate([flatten_res, dense4])

    hidden1 = Dense(512, activation='relu')(merge)
    hidden2 = Dense(1024, activation='relu')(hidden1)
    hidden3 = Dense(2048, activation='relu')(hidden2)
    hidden4 = Dense(2048, activation='relu')(hidden3)
    hidden5 = Dense(512, activation='relu')(hidden4)
    output = Dense(1)(hidden5)

    model = Model(inputs=[visible1, visible2], outputs=output)

    # print(model.summary())
    model.compile(loss='mean_squared_error', optimizer=adam_optimizer)
    model.fit([x_train_img, x_train], y_train,
              validation_data=([x_test_img, x_test], y_test),
              callbacks=[monitor, checkpointer],verbose=2,epochs=150)

print('Training finished...Loading the best model')
print()
model.load_weights(filepath) # load weights from best model

pred_resNet2 = model.predict([x_test_img, x_test])
score_resNet2 = np.sqrt(metrics.mean_squared_error(pred_resNet2,y_test))
print("Score (RMSE): {}".format(score_resNet2))

=======================================  0
Epoch 1/150
8/8 - 23s - 3s/step - loss: 319721209856.0000 - val_loss: 370657591296.0000
Epoch 2/150
8/8 - 2s - 238ms/step - loss: 290786541568.0000 - val_loss: 227072081920.0000
Epoch 3/150
8/8 - 4s - 552ms/step - loss: 118001278976.0000 - val_loss: 29021890560.0000
Epoch 4/150
8/8 - 0s - 55ms/step - loss: 61190516736.0000 - val_loss: 52206501888.0000
Epoch 5/150
8/8 - 0s - 45ms/step - loss: 35417866240.0000 - val_loss: 55066161152.0000
Epoch 6/150
8/8 - 2s - 216ms/step - loss: 32365518848.0000 - val_loss: 27514355712.0000
Epoch 7/150
8/8 - 2s - 215ms/step - loss: 21655685120.0000 - val_loss: 22070456320.0000
Epoch 8/150
8/8 - 2s - 218ms/step - loss: 17482160128.0000 - val_loss: 22067984384.0000
Epoch 9/150
8/8 - 5s - 606ms/step - loss: 15329366016.0000 - val_loss: 20180443136.0000
Epoch 10/150
8/8 - 2s - 282ms/step - loss: 13808812032.0000 - val_loss: 17435619328.0000
Epoch 11/150
8/8 - 0s - 57ms/step - loss: 15548213248.0000 - val_loss: 17636808704.0000
Epoch 12/150
8/8 - 2s - 219ms/step - loss: 10538602496.0000 - val_loss: 13450860544.0000
Epoch 13/150
8/8 - 0s - 43ms/step - loss: 11992697856.0000 - val_loss: 17736079360.0000
Epoch 14/150
8/8 - 0s - 43ms/step - loss: 10125106176.0000 - val_loss: 17297338368.0000
Epoch 15/150
8/8 - 0s - 44ms/step - loss: 9144175616.0000 - val_loss: 17652037632.0000
Epoch 16/150
8/8 - 3s - 397ms/step - loss: 8417947648.0000 - val_loss: 10573316096.0000
Epoch 17/150
8/8 - 0s - 58ms/step - loss: 7398643200.0000 - val_loss: 10875116544.0000
Epoch 18/150
8/8 - 0s - 42ms/step - loss: 6723719168.0000 - val_loss: 11920780288.0000
Epoch 19/150
8/8 - 2s - 218ms/step - loss: 6539856384.0000 - val_loss: 8697222144.0000
Epoch 20/150
8/8 - 0s - 44ms/step - loss: 7597569536.0000 - val_loss: 12927842304.0000
Epoch 21/150
8/8 - 0s - 44ms/step - loss: 8304240640.0000 - val_loss: 12695304192.0000
Epoch 22/150
8/8 - 2s - 272ms/step - loss: 6493758464.0000 - val_loss: 8583385088.0000
Epoch 23/150
8/8 - 0s - 57ms/step - loss: 7047372800.0000 - val_loss: 8963922944.0000
Epoch 24/150
8/8 - 2s - 215ms/step - loss: 5992011776.0000 - val_loss: 8486186496.0000
Epoch 25/150
8/8 - 0s - 44ms/step - loss: 5508170240.0000 - val_loss: 10827868160.0000
Epoch 26/150
8/8 - 0s - 43ms/step - loss: 7335631872.0000 - val_loss: 10224600064.0000
Epoch 27/150
8/8 - 0s - 44ms/step - loss: 7970451968.0000 - val_loss: 9805443072.0000
Epoch 28/150
8/8 - 0s - 45ms/step - loss: 7085118464.0000 - val_loss: 12077138944.0000
Epoch 29/150
8/8 - 0s - 44ms/step - loss: 6370752512.0000 - val_loss: 12534009856.0000
Epoch 30/150
8/8 - 2s - 273ms/step - loss: 6375626240.0000 - val_loss: 7832850944.0000
Epoch 31/150
8/8 - 0s - 54ms/step - loss: 6371791872.0000 - val_loss: 14033116160.0000
Epoch 32/150
8/8 - 0s - 44ms/step - loss: 6926347264.0000 - val_loss: 10506520576.0000
Epoch 33/150
8/8 - 2s - 214ms/step - loss: 4960961536.0000 - val_loss: 7310561792.0000
Epoch 34/150
8/8 - 0s - 44ms/step - loss: 5140932096.0000 - val_loss: 9926811648.0000
Epoch 35/150
8/8 - 0s - 44ms/step - loss: 4800078848.0000 - val_loss: 8254832128.0000
Epoch 36/150
8/8 - 0s - 44ms/step - loss: 4039965952.0000 - val_loss: 7583156736.0000
Epoch 37/150
8/8 - 0s - 44ms/step - loss: 3736108288.0000 - val_loss: 12681845760.0000
Epoch 38/150
8/8 - 0s - 44ms/step - loss: 7528453632.0000 - val_loss: 10006091776.0000
Epoch 39/150
8/8 - 0s - 44ms/step - loss: 5030610432.0000 - val_loss: 15609115648.0000
Epoch 40/150
8/8 - 0s - 44ms/step - loss: 7149642240.0000 - val_loss: 7564073984.0000
Epoch 41/150
8/8 - 0s - 44ms/step - loss: 6850776576.0000 - val_loss: 9907296256.0000
Epoch 42/150
8/8 - 0s - 44ms/step - loss: 7753598464.0000 - val_loss: 10597488640.0000
Epoch 43/150
8/8 - 2s - 226ms/step - loss: 4941890560.0000 - val_loss: 6841288192.0000
Epoch 44/150
8/8 - 0s - 45ms/step - loss: 4671653888.0000 - val_loss: 11344503808.0000
Epoch 45/150
8/8 - 0s - 44ms/step - loss: 4224482304.0000 - val_loss: 7633529856.0000
Epoch 46/150
8/8 - 0s - 44ms/step - loss: 5938935296.0000 - val_loss: 12315461632.0000
Epoch 47/150
8/8 - 0s - 44ms/step - loss: 6503223296.0000 - val_loss: 7215206400.0000
Epoch 48/150
8/8 - 0s - 44ms/step - loss: 8211675648.0000 - val_loss: 6928662528.0000
Epoch 49/150
8/8 - 0s - 44ms/step - loss: 6465079296.0000 - val_loss: 19284856832.0000
Epoch 50/150
8/8 - 0s - 43ms/step - loss: 6603610112.0000 - val_loss: 8105010688.0000
Epoch 51/150
8/8 - 0s - 44ms/step - loss: 4362271232.0000 - val_loss: 8284277248.0000
Epoch 52/150
8/8 - 0s - 44ms/step - loss: 3973266176.0000 - val_loss: 6941509632.0000
Epoch 53/150
8/8 - 0s - 44ms/step - loss: 3667523840.0000 - val_loss: 7579753984.0000
Epoch 54/150
8/8 - 0s - 45ms/step - loss: 2748536576.0000 - val_loss: 7724619776.0000
Epoch 55/150
8/8 - 0s - 44ms/step - loss: 2554678784.0000 - val_loss: 6882731520.0000
Epoch 56/150
8/8 - 0s - 44ms/step - loss: 2334604544.0000 - val_loss: 8000998400.0000
Epoch 57/150
8/8 - 0s - 44ms/step - loss: 3215058688.0000 - val_loss: 9302465536.0000
Epoch 58/150
8/8 - 0s - 46ms/step - loss: 2165706752.0000 - val_loss: 8269378048.0000
Epoch 59/150
8/8 - 2s - 222ms/step - loss: 2756900096.0000 - val_loss: 6685319680.0000
Epoch 60/150
8/8 - 0s - 44ms/step - loss: 4076736768.0000 - val_loss: 7499158528.0000
Epoch 61/150
8/8 - 0s - 44ms/step - loss: 2776761344.0000 - val_loss: 8415473152.0000
Epoch 62/150
8/8 - 0s - 44ms/step - loss: 2841111552.0000 - val_loss: 7466414592.0000
Epoch 63/150
8/8 - 0s - 44ms/step - loss: 2600123136.0000 - val_loss: 8024599040.0000
Epoch 64/150
8/8 - 0s - 44ms/step - loss: 2333960960.0000 - val_loss: 10717296640.0000
Epoch 65/150
8/8 - 0s - 44ms/step - loss: 2490125312.0000 - val_loss: 9897322496.0000
Epoch 66/150
8/8 - 0s - 44ms/step - loss: 2540238080.0000 - val_loss: 7674310656.0000
Epoch 67/150
8/8 - 0s - 45ms/step - loss: 1692400512.0000 - val_loss: 8339127296.0000
Epoch 68/150
8/8 - 0s - 44ms/step - loss: 2159478016.0000 - val_loss: 11031647232.0000
Epoch 69/150
8/8 - 0s - 44ms/step - loss: 2848952064.0000 - val_loss: 14270696448.0000
Epoch 70/150
8/8 - 0s - 44ms/step - loss: 3938407424.0000 - val_loss: 9517571072.0000
Epoch 71/150
8/8 - 0s - 45ms/step - loss: 2908075520.0000 - val_loss: 13247438848.0000
Epoch 72/150
8/8 - 0s - 45ms/step - loss: 3820322816.0000 - val_loss: 10242641920.0000
Epoch 73/150
8/8 - 0s - 45ms/step - loss: 2515527680.0000 - val_loss: 7510495744.0000
Epoch 74/150
8/8 - 0s - 44ms/step - loss: 1700069632.0000 - val_loss: 7690868224.0000
Epoch 75/150
8/8 - 0s - 44ms/step - loss: 2163100672.0000 - val_loss: 8824305664.0000
Epoch 76/150
8/8 - 0s - 44ms/step - loss: 2006841856.0000 - val_loss: 8428301824.0000
Epoch 77/150
8/8 - 0s - 44ms/step - loss: 2896235008.0000 - val_loss: 9930632192.0000
Epoch 78/150
8/8 - 0s - 44ms/step - loss: 2806994944.0000 - val_loss: 13971971072.0000
Epoch 79/150
8/8 - 0s - 44ms/step - loss: 2777547776.0000 - val_loss: 8751472640.0000
Epoch 80/150
8/8 - 0s - 44ms/step - loss: 1549833472.0000 - val_loss: 8155105280.0000
Epoch 81/150
8/8 - 0s - 44ms/step - loss: 1919553280.0000 - val_loss: 8938406912.0000
Epoch 82/150
8/8 - 0s - 44ms/step - loss: 1825622400.0000 - val_loss: 10938185728.0000
Epoch 83/150
8/8 - 0s - 44ms/step - loss: 1716802432.0000 - val_loss: 9716320256.0000
Epoch 84/150
8/8 - 0s - 45ms/step - loss: 2283477504.0000 - val_loss: 10222003200.0000
Epoch 84: early stopping
=======================================  1
Epoch 1/150
8/8 - 23s - 3s/step - loss: 323124264960.0000 - val_loss: 370221547520.0000
Epoch 2/150
8/8 - 0s - 53ms/step - loss: 280841584640.0000 - val_loss: 42515427328.0000
Epoch 3/150
8/8 - 0s - 44ms/step - loss: 48617910272.0000 - val_loss: 55012298752.0000
Epoch 4/150
8/8 - 0s - 44ms/step - loss: 33515941888.0000 - val_loss: 34180034560.0000
Epoch 5/150
8/8 - 0s - 44ms/step - loss: 24238788608.0000 - val_loss: 21874577408.0000
Epoch 6/150
8/8 - 0s - 44ms/step - loss: 19186792448.0000 - val_loss: 22241099776.0000
Epoch 7/150
8/8 - 0s - 44ms/step - loss: 16471012352.0000 - val_loss: 18573068288.0000
Epoch 8/150
8/8 - 0s - 44ms/step - loss: 13675579392.0000 - val_loss: 20579489792.0000
Epoch 9/150
8/8 - 0s - 44ms/step - loss: 13156483072.0000 - val_loss: 23870595072.0000
Epoch 10/150
8/8 - 0s - 44ms/step - loss: 15271145472.0000 - val_loss: 20938483712.0000
Epoch 11/150
8/8 - 0s - 44ms/step - loss: 15053608960.0000 - val_loss: 16290620416.0000
Epoch 12/150
8/8 - 0s - 45ms/step - loss: 12879557632.0000 - val_loss: 23330811904.0000
Epoch 13/150
8/8 - 0s - 44ms/step - loss: 13119328256.0000 - val_loss: 13680312320.0000
Epoch 14/150
8/8 - 0s - 44ms/step - loss: 13158654976.0000 - val_loss: 32319952896.0000
Epoch 15/150
8/8 - 0s - 43ms/step - loss: 19809118208.0000 - val_loss: 22007406592.0000
Epoch 16/150
8/8 - 0s - 46ms/step - loss: 16928901120.0000 - val_loss: 10907711488.0000
Epoch 17/150
8/8 - 0s - 44ms/step - loss: 10646664192.0000 - val_loss: 18891606016.0000
Epoch 18/150
8/8 - 0s - 43ms/step - loss: 9932808192.0000 - val_loss: 10771725312.0000
Epoch 19/150
8/8 - 0s - 44ms/step - loss: 7640130048.0000 - val_loss: 9787567104.0000
Epoch 20/150
8/8 - 0s - 44ms/step - loss: 6777099264.0000 - val_loss: 14609636352.0000
Epoch 21/150
8/8 - 0s - 44ms/step - loss: 7358688768.0000 - val_loss: 9857030144.0000
Epoch 22/150
8/8 - 0s - 43ms/step - loss: 7909317632.0000 - val_loss: 13521220608.0000
Epoch 23/150
8/8 - 0s - 44ms/step - loss: 6962408960.0000 - val_loss: 9131530240.0000
Epoch 24/150
8/8 - 0s - 44ms/step - loss: 6194423296.0000 - val_loss: 10061654016.0000
Epoch 25/150
8/8 - 0s - 44ms/step - loss: 5581868544.0000 - val_loss: 10461538304.0000
Epoch 26/150
8/8 - 0s - 44ms/step - loss: 6399563776.0000 - val_loss: 10648406016.0000
Epoch 27/150
8/8 - 0s - 44ms/step - loss: 5708452864.0000 - val_loss: 11247179776.0000
Epoch 28/150
8/8 - 0s - 44ms/step - loss: 8574018048.0000 - val_loss: 9227262976.0000
Epoch 29/150
8/8 - 0s - 44ms/step - loss: 6712791552.0000 - val_loss: 8725606400.0000
Epoch 30/150
8/8 - 0s - 44ms/step - loss: 4667999232.0000 - val_loss: 11859616768.0000
Epoch 31/150
8/8 - 0s - 44ms/step - loss: 5393193472.0000 - val_loss: 9648734208.0000
Epoch 32/150
8/8 - 0s - 44ms/step - loss: 6554749952.0000 - val_loss: 12750115840.0000
Epoch 33/150
8/8 - 0s - 44ms/step - loss: 6426609152.0000 - val_loss: 13898280960.0000
Epoch 34/150
8/8 - 0s - 44ms/step - loss: 5829546496.0000 - val_loss: 10843250688.0000
Epoch 35/150
8/8 - 0s - 44ms/step - loss: 5386816000.0000 - val_loss: 14843464704.0000
Epoch 36/150
8/8 - 0s - 44ms/step - loss: 6304473600.0000 - val_loss: 9510558720.0000
Epoch 37/150
8/8 - 0s - 45ms/step - loss: 6006458368.0000 - val_loss: 15671720960.0000
Epoch 38/150
8/8 - 0s - 44ms/step - loss: 7293254656.0000 - val_loss: 9463809024.0000
Epoch 39/150
8/8 - 0s - 44ms/step - loss: 4516019712.0000 - val_loss: 9224008704.0000
Epoch 40/150
8/8 - 0s - 44ms/step - loss: 4288998656.0000 - val_loss: 9594572800.0000
Epoch 41/150
8/8 - 0s - 44ms/step - loss: 4333754368.0000 - val_loss: 10188886016.0000
Epoch 42/150
8/8 - 0s - 44ms/step - loss: 4284501248.0000 - val_loss: 9757449216.0000
Epoch 43/150
8/8 - 0s - 45ms/step - loss: 4135467776.0000 - val_loss: 14799822848.0000
Epoch 44/150
8/8 - 0s - 46ms/step - loss: 5738757120.0000 - val_loss: 17976887296.0000
Epoch 45/150
8/8 - 0s - 44ms/step - loss: 8246518784.0000 - val_loss: 19484778496.0000
Epoch 46/150
8/8 - 0s - 44ms/step - loss: 7760026112.0000 - val_loss: 10313184256.0000
Epoch 47/150
8/8 - 0s - 44ms/step - loss: 5194813952.0000 - val_loss: 11470895104.0000
Epoch 48/150
8/8 - 0s - 44ms/step - loss: 5281802752.0000 - val_loss: 9331975168.0000
Epoch 49/150
8/8 - 0s - 45ms/step - loss: 4152193536.0000 - val_loss: 9266804736.0000
Epoch 50/150
8/8 - 0s - 44ms/step - loss: 5914046976.0000 - val_loss: 13175555072.0000
Epoch 51/150
8/8 - 0s - 44ms/step - loss: 5872590336.0000 - val_loss: 10324704256.0000
Epoch 52/150
8/8 - 0s - 44ms/step - loss: 5051201024.0000 - val_loss: 13209012224.0000
Epoch 53/150
8/8 - 0s - 45ms/step - loss: 4757768192.0000 - val_loss: 12152287232.0000
Epoch 54/150
8/8 - 0s - 45ms/step - loss: 4264729856.0000 - val_loss: 11725083648.0000
Epoch 54: early stopping
=======================================  2
Epoch 1/150
8/8 - 21s - 3s/step - loss: 344623808512.0000 - val_loss: 370955124736.0000
Epoch 2/150
8/8 - 0s - 54ms/step - loss: 546286370816.0000 - val_loss: 224578224128.0000
Epoch 3/150
8/8 - 0s - 43ms/step - loss: 263560888320.0000 - val_loss: 365780303872.0000
Epoch 4/150
8/8 - 0s - 44ms/step - loss: 354419146752.0000 - val_loss: 366570995712.0000
Epoch 5/150
8/8 - 0s - 43ms/step - loss: 311703764992.0000 - val_loss: 328785526784.0000
Epoch 6/150
8/8 - 0s - 43ms/step - loss: 153192284160.0000 - val_loss: 191294144512.0000
Epoch 7/150
8/8 - 0s - 44ms/step - loss: 66382794752.0000 - val_loss: 71911161856.0000
Epoch 8/150
8/8 - 0s - 44ms/step - loss: 43769565184.0000 - val_loss: 73560678400.0000
Epoch 9/150
8/8 - 0s - 44ms/step - loss: 32644128768.0000 - val_loss: 35337408512.0000
Epoch 10/150
8/8 - 0s - 44ms/step - loss: 19179958272.0000 - val_loss: 28929382400.0000
Epoch 11/150
8/8 - 0s - 44ms/step - loss: 13173669888.0000 - val_loss: 28569276416.0000
Epoch 12/150
8/8 - 0s - 46ms/step - loss: 12533525504.0000 - val_loss: 24265093120.0000
Epoch 13/150
8/8 - 0s - 44ms/step - loss: 11787664384.0000 - val_loss: 23159500800.0000
Epoch 14/150
8/8 - 0s - 43ms/step - loss: 10243992576.0000 - val_loss: 21414912000.0000
Epoch 15/150
8/8 - 0s - 43ms/step - loss: 10104990720.0000 - val_loss: 22558169088.0000
Epoch 16/150
8/8 - 0s - 43ms/step - loss: 10711566336.0000 - val_loss: 25117384704.0000
Epoch 17/150
8/8 - 0s - 44ms/step - loss: 10217789440.0000 - val_loss: 20548102144.0000
Epoch 18/150
8/8 - 0s - 43ms/step - loss: 8842452992.0000 - val_loss: 23685742592.0000
Epoch 19/150
8/8 - 0s - 44ms/step - loss: 8758758400.0000 - val_loss: 18865295360.0000
Epoch 20/150
8/8 - 0s - 44ms/step - loss: 7914476544.0000 - val_loss: 18802552832.0000
Epoch 21/150
8/8 - 0s - 44ms/step - loss: 9545149440.0000 - val_loss: 19648262144.0000
Epoch 22/150
8/8 - 0s - 43ms/step - loss: 7907062272.0000 - val_loss: 17209364480.0000
Epoch 23/150
8/8 - 0s - 44ms/step - loss: 7948295680.0000 - val_loss: 19794915328.0000
Epoch 24/150
8/8 - 0s - 44ms/step - loss: 7782191104.0000 - val_loss: 18811815936.0000
Epoch 25/150
8/8 - 0s - 44ms/step - loss: 6909862400.0000 - val_loss: 20675829760.0000
Epoch 26/150
8/8 - 0s - 44ms/step - loss: 7292512768.0000 - val_loss: 19528013824.0000
Epoch 27/150
8/8 - 0s - 44ms/step - loss: 6844777984.0000 - val_loss: 16698336256.0000
Epoch 28/150
8/8 - 0s - 44ms/step - loss: 6874917888.0000 - val_loss: 17318799360.0000
Epoch 29/150
8/8 - 0s - 44ms/step - loss: 7401374720.0000 - val_loss: 18347407360.0000
Epoch 30/150
8/8 - 0s - 44ms/step - loss: 6420854784.0000 - val_loss: 18002798592.0000
Epoch 31/150
8/8 - 0s - 44ms/step - loss: 6393824768.0000 - val_loss: 17272932352.0000
Epoch 32/150
8/8 - 0s - 44ms/step - loss: 6444038144.0000 - val_loss: 16520418304.0000
Epoch 33/150
8/8 - 0s - 45ms/step - loss: 6893662720.0000 - val_loss: 18202013696.0000
Epoch 34/150
8/8 - 0s - 47ms/step - loss: 6096175104.0000 - val_loss: 17366958080.0000
Epoch 35/150
8/8 - 0s - 44ms/step - loss: 7121977344.0000 - val_loss: 17174304768.0000
Epoch 36/150
8/8 - 0s - 45ms/step - loss: 5943210496.0000 - val_loss: 17019105280.0000
Epoch 37/150
8/8 - 0s - 44ms/step - loss: 6029882880.0000 - val_loss: 16938865664.0000
Epoch 38/150
8/8 - 0s - 44ms/step - loss: 5968013824.0000 - val_loss: 19029760000.0000
Epoch 39/150
8/8 - 0s - 44ms/step - loss: 8530981888.0000 - val_loss: 21932326912.0000
Epoch 40/150
8/8 - 0s - 45ms/step - loss: 8958773248.0000 - val_loss: 19374319616.0000
Epoch 41/150
8/8 - 0s - 44ms/step - loss: 8776313856.0000 - val_loss: 17294534656.0000
Epoch 42/150
8/8 - 0s - 44ms/step - loss: 7528290816.0000 - val_loss: 18980823040.0000
Epoch 43/150
8/8 - 0s - 44ms/step - loss: 6767010816.0000 - val_loss: 22390620160.0000
Epoch 44/150
8/8 - 0s - 44ms/step - loss: 6701046784.0000 - val_loss: 18354167808.0000
Epoch 45/150
8/8 - 0s - 44ms/step - loss: 7716420608.0000 - val_loss: 17131030528.0000
Epoch 46/150
8/8 - 0s - 44ms/step - loss: 5908751872.0000 - val_loss: 18244859904.0000
Epoch 47/150
8/8 - 0s - 44ms/step - loss: 6926004736.0000 - val_loss: 20134727680.0000
Epoch 48/150
8/8 - 0s - 44ms/step - loss: 6723584512.0000 - val_loss: 19048880128.0000
Epoch 49/150
8/8 - 0s - 44ms/step - loss: 5545666048.0000 - val_loss: 17418766336.0000
Epoch 50/150
8/8 - 0s - 44ms/step - loss: 5137236480.0000 - val_loss: 16449628160.0000
Epoch 51/150
8/8 - 0s - 44ms/step - loss: 5013325312.0000 - val_loss: 23336169472.0000
Epoch 52/150
8/8 - 0s - 44ms/step - loss: 6988819968.0000 - val_loss: 18803173376.0000
Epoch 53/150
8/8 - 0s - 44ms/step - loss: 6073098752.0000 - val_loss: 20673880064.0000
Epoch 54/150
8/8 - 0s - 44ms/step - loss: 6946417152.0000 - val_loss: 19777730560.0000
Epoch 55/150
8/8 - 0s - 44ms/step - loss: 5307472896.0000 - val_loss: 17794631680.0000
Epoch 56/150
8/8 - 0s - 44ms/step - loss: 4963123712.0000 - val_loss: 17786304512.0000
Epoch 57/150
8/8 - 0s - 44ms/step - loss: 4891612672.0000 - val_loss: 17979236352.0000
Epoch 58/150
8/8 - 0s - 44ms/step - loss: 4558477312.0000 - val_loss: 17798459392.0000
Epoch 59/150
8/8 - 0s - 44ms/step - loss: 4801461248.0000 - val_loss: 18605156352.0000
Epoch 60/150
8/8 - 0s - 44ms/step - loss: 4632326144.0000 - val_loss: 18636691456.0000
Epoch 61/150
8/8 - 0s - 44ms/step - loss: 7221147648.0000 - val_loss: 18749089792.0000
Epoch 62/150
8/8 - 0s - 44ms/step - loss: 6871911424.0000 - val_loss: 18182045696.0000
Epoch 63/150
8/8 - 0s - 44ms/step - loss: 5877418496.0000 - val_loss: 18186409984.0000
Epoch 64/150
8/8 - 0s - 44ms/step - loss: 5621193728.0000 - val_loss: 24608532480.0000
Epoch 65/150
8/8 - 0s - 44ms/step - loss: 6892570624.0000 - val_loss: 19988211712.0000
Epoch 66/150
8/8 - 0s - 44ms/step - loss: 5883388416.0000 - val_loss: 19565705216.0000
Epoch 67/150
8/8 - 0s - 44ms/step - loss: 5317914624.0000 - val_loss: 19195856896.0000
Epoch 68/150
8/8 - 0s - 45ms/step - loss: 5492926976.0000 - val_loss: 18788526080.0000
Epoch 69/150
8/8 - 0s - 44ms/step - loss: 5800559104.0000 - val_loss: 18961080320.0000
Epoch 70/150
8/8 - 0s - 44ms/step - loss: 4771607552.0000 - val_loss: 18856216576.0000
Epoch 71/150
8/8 - 0s - 44ms/step - loss: 4740642816.0000 - val_loss: 19100248064.0000
Epoch 72/150
8/8 - 0s - 44ms/step - loss: 4538798592.0000 - val_loss: 18112870400.0000
Epoch 73/150
8/8 - 0s - 44ms/step - loss: 5075757056.0000 - val_loss: 21610674176.0000
Epoch 74/150
8/8 - 0s - 44ms/step - loss: 5952310784.0000 - val_loss: 19308371968.0000
Epoch 75/150
8/8 - 0s - 44ms/step - loss: 5342662656.0000 - val_loss: 18504763392.0000
Epoch 75: early stopping
=======================================  3
Epoch 1/150
8/8 - 21s - 3s/step - loss: 333131808768.0000 - val_loss: 370968002560.0000
Epoch 2/150
8/8 - 0s - 56ms/step - loss: 343679827968.0000 - val_loss: 129755037696.0000
Epoch 3/150
8/8 - 0s - 43ms/step - loss: 206880374784.0000 - val_loss: 239132622848.0000
Epoch 4/150
8/8 - 0s - 43ms/step - loss: 141423919104.0000 - val_loss: 53438222336.0000
Epoch 5/150
8/8 - 0s - 44ms/step - loss: 132658012160.0000 - val_loss: 51341033472.0000
Epoch 6/150
8/8 - 0s - 44ms/step - loss: 65579847680.0000 - val_loss: 74466467840.0000
Epoch 7/150
8/8 - 0s - 44ms/step - loss: 31912544256.0000 - val_loss: 44194017280.0000
Epoch 8/150
8/8 - 0s - 43ms/step - loss: 22672590848.0000 - val_loss: 30234505216.0000
Epoch 9/150
8/8 - 0s - 44ms/step - loss: 16969053184.0000 - val_loss: 25548308480.0000
Epoch 10/150
8/8 - 0s - 43ms/step - loss: 15006684160.0000 - val_loss: 22143371264.0000
Epoch 11/150
8/8 - 0s - 43ms/step - loss: 12730932224.0000 - val_loss: 22571356160.0000
Epoch 12/150
8/8 - 0s - 43ms/step - loss: 12437800960.0000 - val_loss: 19680249856.0000
Epoch 13/150
8/8 - 0s - 43ms/step - loss: 12513068032.0000 - val_loss: 22539323392.0000
Epoch 14/150
8/8 - 0s - 45ms/step - loss: 12232555520.0000 - val_loss: 20198987776.0000
Epoch 15/150
8/8 - 0s - 44ms/step - loss: 10190428160.0000 - val_loss: 20771207168.0000
Epoch 16/150
8/8 - 0s - 43ms/step - loss: 11294713856.0000 - val_loss: 16391320576.0000
Epoch 17/150
8/8 - 0s - 44ms/step - loss: 9611409408.0000 - val_loss: 16413604864.0000
Epoch 18/150
8/8 - 0s - 43ms/step - loss: 10863264768.0000 - val_loss: 20356630528.0000
Epoch 19/150
8/8 - 0s - 44ms/step - loss: 9911248896.0000 - val_loss: 20044722176.0000
Epoch 20/150
8/8 - 0s - 44ms/step - loss: 10130038784.0000 - val_loss: 17788657664.0000
Epoch 21/150
8/8 - 0s - 43ms/step - loss: 10617415680.0000 - val_loss: 17314623488.0000
Epoch 22/150
8/8 - 0s - 44ms/step - loss: 9966184448.0000 - val_loss: 18128560128.0000
Epoch 23/150
8/8 - 0s - 43ms/step - loss: 10636024832.0000 - val_loss: 18197403648.0000
Epoch 24/150
8/8 - 0s - 44ms/step - loss: 9380912128.0000 - val_loss: 15288996864.0000
Epoch 25/150
8/8 - 0s - 44ms/step - loss: 8787020800.0000 - val_loss: 18786781184.0000
Epoch 26/150
8/8 - 0s - 44ms/step - loss: 10612093952.0000 - val_loss: 15099997184.0000
Epoch 27/150
8/8 - 0s - 44ms/step - loss: 12092460032.0000 - val_loss: 20376549376.0000
Epoch 28/150
8/8 - 0s - 44ms/step - loss: 9618940928.0000 - val_loss: 18043412480.0000
Epoch 29/150
8/8 - 0s - 44ms/step - loss: 7692622848.0000 - val_loss: 15368712192.0000
Epoch 30/150
8/8 - 0s - 44ms/step - loss: 7563075072.0000 - val_loss: 15547862016.0000
Epoch 31/150
8/8 - 0s - 44ms/step - loss: 7652174848.0000 - val_loss: 15095064576.0000
Epoch 32/150
8/8 - 0s - 44ms/step - loss: 9178969088.0000 - val_loss: 20978319360.0000
Epoch 33/150
8/8 - 0s - 44ms/step - loss: 8956332032.0000 - val_loss: 25532321792.0000
Epoch 34/150
8/8 - 0s - 44ms/step - loss: 11470585856.0000 - val_loss: 15840988160.0000
Epoch 35/150
8/8 - 0s - 44ms/step - loss: 7925054976.0000 - val_loss: 16660546560.0000
Epoch 36/150
8/8 - 0s - 44ms/step - loss: 8721024000.0000 - val_loss: 15629094912.0000
Epoch 37/150
8/8 - 0s - 43ms/step - loss: 8928829440.0000 - val_loss: 14865820672.0000
Epoch 38/150
8/8 - 0s - 44ms/step - loss: 6986847232.0000 - val_loss: 14541087744.0000
Epoch 39/150
8/8 - 0s - 44ms/step - loss: 7957356544.0000 - val_loss: 15079907328.0000
Epoch 40/150
8/8 - 0s - 44ms/step - loss: 6132814848.0000 - val_loss: 19793524736.0000
Epoch 41/150
8/8 - 0s - 44ms/step - loss: 9682405376.0000 - val_loss: 15511847936.0000
Epoch 42/150
8/8 - 0s - 45ms/step - loss: 7938184704.0000 - val_loss: 15161032704.0000
Epoch 43/150
8/8 - 0s - 44ms/step - loss: 11871593472.0000 - val_loss: 22093836288.0000
Epoch 44/150
8/8 - 0s - 44ms/step - loss: 10286960640.0000 - val_loss: 18485227520.0000
Epoch 45/150
8/8 - 0s - 44ms/step - loss: 14832917504.0000 - val_loss: 29872973824.0000
Epoch 46/150
8/8 - 0s - 44ms/step - loss: 13397815296.0000 - val_loss: 20867938304.0000
Epoch 47/150
8/8 - 0s - 43ms/step - loss: 10589310976.0000 - val_loss: 19198025728.0000
Epoch 48/150
8/8 - 0s - 44ms/step - loss: 9587639296.0000 - val_loss: 17917896704.0000
Epoch 49/150
8/8 - 0s - 44ms/step - loss: 8316842496.0000 - val_loss: 15841428480.0000
Epoch 50/150
8/8 - 0s - 44ms/step - loss: 7685756928.0000 - val_loss: 15254457344.0000
Epoch 51/150
8/8 - 0s - 44ms/step - loss: 7522620416.0000 - val_loss: 18847922176.0000
Epoch 52/150
8/8 - 0s - 44ms/step - loss: 7552425472.0000 - val_loss: 16646581248.0000
Epoch 53/150
8/8 - 0s - 44ms/step - loss: 10177015808.0000 - val_loss: 18371600384.0000
Epoch 54/150
8/8 - 0s - 44ms/step - loss: 8900449280.0000 - val_loss: 15992401920.0000
Epoch 55/150
8/8 - 0s - 44ms/step - loss: 6246259200.0000 - val_loss: 14747586560.0000
Epoch 56/150
8/8 - 0s - 44ms/step - loss: 6015745536.0000 - val_loss: 14402470912.0000
Epoch 57/150
8/8 - 0s - 44ms/step - loss: 6469605376.0000 - val_loss: 17437169664.0000
Epoch 58/150
8/8 - 0s - 44ms/step - loss: 8129166336.0000 - val_loss: 16658107392.0000
Epoch 59/150
8/8 - 0s - 44ms/step - loss: 8069288960.0000 - val_loss: 15207404544.0000
Epoch 60/150
8/8 - 0s - 44ms/step - loss: 9192803328.0000 - val_loss: 24142895104.0000
Epoch 61/150
8/8 - 0s - 44ms/step - loss: 8063428608.0000 - val_loss: 14136548352.0000
Epoch 62/150
8/8 - 0s - 44ms/step - loss: 6224045568.0000 - val_loss: 17265704960.0000
Epoch 63/150
8/8 - 0s - 44ms/step - loss: 6521044992.0000 - val_loss: 13302090752.0000
Epoch 64/150
8/8 - 0s - 44ms/step - loss: 6801262080.0000 - val_loss: 19809384448.0000
Epoch 65/150
8/8 - 0s - 44ms/step - loss: 8537410048.0000 - val_loss: 19630039040.0000
Epoch 66/150
8/8 - 0s - 44ms/step - loss: 6393229824.0000 - val_loss: 16643032064.0000
Epoch 67/150
8/8 - 0s - 44ms/step - loss: 5234432000.0000 - val_loss: 17820147712.0000
Epoch 68/150
8/8 - 0s - 44ms/step - loss: 8140945408.0000 - val_loss: 15458912256.0000
Epoch 69/150
8/8 - 0s - 44ms/step - loss: 8007631872.0000 - val_loss: 15876874240.0000
Epoch 70/150
8/8 - 0s - 45ms/step - loss: 6373708288.0000 - val_loss: 18539823104.0000
Epoch 71/150
8/8 - 0s - 44ms/step - loss: 6840059392.0000 - val_loss: 17326911488.0000
Epoch 72/150
8/8 - 0s - 46ms/step - loss: 10346519552.0000 - val_loss: 18447204352.0000
Epoch 73/150
8/8 - 0s - 47ms/step - loss: 7819506688.0000 - val_loss: 14696387584.0000
Epoch 74/150
8/8 - 0s - 45ms/step - loss: 6052153344.0000 - val_loss: 21267703808.0000
Epoch 75/150
8/8 - 0s - 45ms/step - loss: 7956389888.0000 - val_loss: 16388240384.0000
Epoch 76/150
8/8 - 0s - 45ms/step - loss: 7196965888.0000 - val_loss: 20423415808.0000
Epoch 77/150
8/8 - 0s - 44ms/step - loss: 5733415936.0000 - val_loss: 16421949440.0000
Epoch 78/150
8/8 - 0s - 44ms/step - loss: 7321218048.0000 - val_loss: 18719490048.0000
Epoch 79/150
8/8 - 0s - 44ms/step - loss: 7028423680.0000 - val_loss: 15059024896.0000
Epoch 80/150
8/8 - 0s - 44ms/step - loss: 5440313344.0000 - val_loss: 16120580096.0000
Epoch 81/150
8/8 - 0s - 44ms/step - loss: 4762501120.0000 - val_loss: 16071350272.0000
Epoch 82/150
8/8 - 0s - 44ms/step - loss: 5238790144.0000 - val_loss: 14154473472.0000
Epoch 83/150
8/8 - 0s - 44ms/step - loss: 5607762432.0000 - val_loss: 15561421824.0000
Epoch 84/150
8/8 - 0s - 44ms/step - loss: 4920118272.0000 - val_loss: 15625426944.0000
Epoch 85/150
8/8 - 0s - 44ms/step - loss: 5326107136.0000 - val_loss: 15925782528.0000
Epoch 86/150
8/8 - 0s - 45ms/step - loss: 4760399872.0000 - val_loss: 18455529472.0000
Epoch 87/150
8/8 - 0s - 44ms/step - loss: 5973572096.0000 - val_loss: 14240203776.0000
Epoch 88/150
8/8 - 0s - 44ms/step - loss: 5464611328.0000 - val_loss: 15258218496.0000
Epoch 88: early stopping
=======================================  4
Epoch 1/150
8/8 - 23s - 3s/step - loss: 332965412864.0000 - val_loss: 370958860288.0000
Epoch 2/150
8/8 - 0s - 55ms/step - loss: 311689969664.0000 - val_loss: 843722391552.0000
Epoch 3/150
8/8 - 0s - 44ms/step - loss: 183587209216.0000 - val_loss: 164074782720.0000
Epoch 4/150
8/8 - 0s - 43ms/step - loss: 105109323776.0000 - val_loss: 130946908160.0000
Epoch 5/150
8/8 - 0s - 46ms/step - loss: 70926548992.0000 - val_loss: 57204752384.0000
Epoch 6/150
8/8 - 0s - 44ms/step - loss: 38607491072.0000 - val_loss: 24503250944.0000
Epoch 7/150
8/8 - 0s - 44ms/step - loss: 34014586880.0000 - val_loss: 25708242944.0000
Epoch 8/150
8/8 - 0s - 44ms/step - loss: 24708632576.0000 - val_loss: 21385943040.0000
Epoch 9/150
8/8 - 0s - 45ms/step - loss: 21538289664.0000 - val_loss: 21273319424.0000
Epoch 10/150
8/8 - 0s - 44ms/step - loss: 17793763328.0000 - val_loss: 20731068416.0000
Epoch 11/150
8/8 - 0s - 44ms/step - loss: 14506426368.0000 - val_loss: 19487899648.0000
Epoch 12/150
8/8 - 0s - 44ms/step - loss: 13694493696.0000 - val_loss: 19452102656.0000
Epoch 13/150
8/8 - 0s - 44ms/step - loss: 12544518144.0000 - val_loss: 24227569664.0000
Epoch 14/150
8/8 - 0s - 44ms/step - loss: 11369886720.0000 - val_loss: 21263362048.0000
Epoch 15/150
8/8 - 0s - 44ms/step - loss: 8978388992.0000 - val_loss: 19227746304.0000
Epoch 16/150
8/8 - 1s - 77ms/step - loss: 9172440064.0000 - val_loss: 18490136576.0000
Epoch 17/150
8/8 - 0s - 44ms/step - loss: 7925522432.0000 - val_loss: 16481592320.0000
Epoch 18/150
8/8 - 0s - 44ms/step - loss: 8106087936.0000 - val_loss: 16990749696.0000
Epoch 19/150
8/8 - 0s - 44ms/step - loss: 10041021440.0000 - val_loss: 25744300032.0000
Epoch 20/150
8/8 - 0s - 44ms/step - loss: 9144858624.0000 - val_loss: 14402652160.0000
Epoch 21/150
8/8 - 0s - 44ms/step - loss: 9329552384.0000 - val_loss: 15419848704.0000
Epoch 22/150
8/8 - 0s - 43ms/step - loss: 7650539520.0000 - val_loss: 15872695296.0000
Epoch 23/150
8/8 - 0s - 44ms/step - loss: 7171934208.0000 - val_loss: 14374939648.0000
Epoch 24/150
8/8 - 0s - 44ms/step - loss: 7069203456.0000 - val_loss: 15751160832.0000
Epoch 25/150
8/8 - 0s - 46ms/step - loss: 7716983296.0000 - val_loss: 16544926720.0000
Epoch 26/150
8/8 - 0s - 44ms/step - loss: 8064707584.0000 - val_loss: 14972206080.0000
Epoch 27/150
8/8 - 0s - 43ms/step - loss: 8233238528.0000 - val_loss: 17242052608.0000
Epoch 28/150
8/8 - 0s - 44ms/step - loss: 7033746944.0000 - val_loss: 14383555584.0000
Epoch 29/150
8/8 - 0s - 44ms/step - loss: 7389596160.0000 - val_loss: 16510401536.0000
Epoch 30/150
8/8 - 0s - 44ms/step - loss: 8516790272.0000 - val_loss: 19883059200.0000
Epoch 31/150
8/8 - 0s - 43ms/step - loss: 10686635008.0000 - val_loss: 15531506688.0000
Epoch 32/150
8/8 - 0s - 44ms/step - loss: 7377322496.0000 - val_loss: 16119406592.0000
Epoch 33/150
8/8 - 0s - 44ms/step - loss: 9107068928.0000 - val_loss: 15687796736.0000
Epoch 34/150
8/8 - 0s - 44ms/step - loss: 7016641024.0000 - val_loss: 15215364096.0000
Epoch 35/150
8/8 - 0s - 44ms/step - loss: 6406585856.0000 - val_loss: 17459034112.0000
Epoch 36/150
8/8 - 0s - 44ms/step - loss: 7242393088.0000 - val_loss: 15672602624.0000
Epoch 37/150
8/8 - 0s - 44ms/step - loss: 7794547712.0000 - val_loss: 14953553920.0000
Epoch 38/150
8/8 - 0s - 44ms/step - loss: 7043517440.0000 - val_loss: 16215073792.0000
Epoch 39/150
8/8 - 0s - 44ms/step - loss: 6555794944.0000 - val_loss: 13372609536.0000
Epoch 40/150
8/8 - 0s - 44ms/step - loss: 6273844736.0000 - val_loss: 19161835520.0000
Epoch 41/150
8/8 - 0s - 44ms/step - loss: 8016130048.0000 - val_loss: 15376745472.0000
Epoch 42/150
8/8 - 0s - 45ms/step - loss: 6483491328.0000 - val_loss: 12808914944.0000
Epoch 43/150
8/8 - 0s - 44ms/step - loss: 6627310592.0000 - val_loss: 16744270848.0000
Epoch 44/150
8/8 - 0s - 44ms/step - loss: 5979522048.0000 - val_loss: 16069594112.0000
Epoch 45/150
8/8 - 0s - 45ms/step - loss: 7897920000.0000 - val_loss: 21201719296.0000
Epoch 46/150
8/8 - 0s - 44ms/step - loss: 8547323392.0000 - val_loss: 18554476544.0000
Epoch 47/150
8/8 - 0s - 44ms/step - loss: 8024240128.0000 - val_loss: 21111246848.0000
Epoch 48/150
8/8 - 0s - 44ms/step - loss: 9059536896.0000 - val_loss: 14481491968.0000
Epoch 49/150
8/8 - 0s - 44ms/step - loss: 7290400768.0000 - val_loss: 16314216448.0000
Epoch 50/150
8/8 - 0s - 44ms/step - loss: 7340284416.0000 - val_loss: 16863049728.0000
Epoch 51/150
8/8 - 0s - 44ms/step - loss: 6321382400.0000 - val_loss: 13021291520.0000
Epoch 52/150
8/8 - 0s - 44ms/step - loss: 6172838400.0000 - val_loss: 14301253632.0000
Epoch 53/150
8/8 - 0s - 46ms/step - loss: 6840600064.0000 - val_loss: 18244388864.0000
Epoch 54/150
8/8 - 0s - 45ms/step - loss: 8305925632.0000 - val_loss: 14192798720.0000
Epoch 55/150
8/8 - 0s - 44ms/step - loss: 8396320768.0000 - val_loss: 15081013248.0000
Epoch 56/150
8/8 - 0s - 44ms/step - loss: 6818115584.0000 - val_loss: 18243239936.0000
Epoch 57/150
8/8 - 0s - 44ms/step - loss: 6558855168.0000 - val_loss: 18747217920.0000
Epoch 58/150
8/8 - 0s - 44ms/step - loss: 6120394240.0000 - val_loss: 19211184128.0000
Epoch 59/150
8/8 - 0s - 44ms/step - loss: 10241474560.0000 - val_loss: 16295349248.0000
Epoch 60/150
8/8 - 0s - 44ms/step - loss: 6168457728.0000 - val_loss: 16996061184.0000
Epoch 61/150
8/8 - 0s - 44ms/step - loss: 5678909440.0000 - val_loss: 14814504960.0000
Epoch 62/150
8/8 - 0s - 44ms/step - loss: 5442418688.0000 - val_loss: 27306475520.0000
Epoch 63/150
8/8 - 0s - 44ms/step - loss: 10215317504.0000 - val_loss: 20827006976.0000
Epoch 64/150
8/8 - 0s - 44ms/step - loss: 10403002368.0000 - val_loss: 15504002048.0000
Epoch 65/150
8/8 - 0s - 44ms/step - loss: 5530212352.0000 - val_loss: 17414854656.0000
Epoch 66/150
8/8 - 0s - 44ms/step - loss: 5985874944.0000 - val_loss: 20766240768.0000
Epoch 67/150
8/8 - 0s - 45ms/step - loss: 8893236224.0000 - val_loss: 29824243712.0000
Epoch 67: early stopping
Training finished...Loading the best model

2/2 ━━━━━━━━━━━━━━━━━━━━ 7s 3s/step
Score (RMSE): 81763.8046875
In [84]:
chart_regression(pred_resNet2.flatten(),y_test)
No description has been provided for this image
In [85]:
print(model.summary())
plot_model(model, show_shapes=True)

Model: "functional_66"

┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓
┃ Layer (type)        ┃ Output Shape      ┃    Param # ┃ Connected to      ┃
┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩
│ input_layer_67      │ (None, 128, 128,  │          0 │ -                 │
│ (InputLayer)        │ 3)                │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv1_pad           │ (None, 134, 134,  │          0 │ input_layer_67[0… │
│ (ZeroPadding2D)     │ 3)                │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv1_conv (Conv2D) │ (None, 64, 64,    │      9,472 │ conv1_pad[0][0]   │
│                     │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv1_bn            │ (None, 64, 64,    │        256 │ conv1_conv[0][0]  │
│ (BatchNormalizatio… │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv1_relu          │ (None, 64, 64,    │          0 │ conv1_bn[0][0]    │
│ (Activation)        │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ pool1_pad           │ (None, 66, 66,    │          0 │ conv1_relu[0][0]  │
│ (ZeroPadding2D)     │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ pool1_pool          │ (None, 32, 32,    │          0 │ pool1_pad[0][0]   │
│ (MaxPooling2D)      │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_1_conv │ (None, 32, 32,    │      4,160 │ pool1_pool[0][0]  │
│ (Conv2D)            │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_1_bn   │ (None, 32, 32,    │        256 │ conv2_block1_1_c… │
│ (BatchNormalizatio… │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_1_relu │ (None, 32, 32,    │          0 │ conv2_block1_1_b… │
│ (Activation)        │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_2_conv │ (None, 32, 32,    │     36,928 │ conv2_block1_1_r… │
│ (Conv2D)            │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_2_bn   │ (None, 32, 32,    │        256 │ conv2_block1_2_c… │
│ (BatchNormalizatio… │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_2_relu │ (None, 32, 32,    │          0 │ conv2_block1_2_b… │
│ (Activation)        │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_0_conv │ (None, 32, 32,    │     16,640 │ pool1_pool[0][0]  │
│ (Conv2D)            │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_3_conv │ (None, 32, 32,    │     16,640 │ conv2_block1_2_r… │
│ (Conv2D)            │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_0_bn   │ (None, 32, 32,    │      1,024 │ conv2_block1_0_c… │
│ (BatchNormalizatio… │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_3_bn   │ (None, 32, 32,    │      1,024 │ conv2_block1_3_c… │
│ (BatchNormalizatio… │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_add    │ (None, 32, 32,    │          0 │ conv2_block1_0_b… │
│ (Add)               │ 256)              │            │ conv2_block1_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block1_out    │ (None, 32, 32,    │          0 │ conv2_block1_add… │
│ (Activation)        │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_1_conv │ (None, 32, 32,    │     16,448 │ conv2_block1_out… │
│ (Conv2D)            │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_1_bn   │ (None, 32, 32,    │        256 │ conv2_block2_1_c… │
│ (BatchNormalizatio… │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_1_relu │ (None, 32, 32,    │          0 │ conv2_block2_1_b… │
│ (Activation)        │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_2_conv │ (None, 32, 32,    │     36,928 │ conv2_block2_1_r… │
│ (Conv2D)            │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_2_bn   │ (None, 32, 32,    │        256 │ conv2_block2_2_c… │
│ (BatchNormalizatio… │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_2_relu │ (None, 32, 32,    │          0 │ conv2_block2_2_b… │
│ (Activation)        │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_3_conv │ (None, 32, 32,    │     16,640 │ conv2_block2_2_r… │
│ (Conv2D)            │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_3_bn   │ (None, 32, 32,    │      1,024 │ conv2_block2_3_c… │
│ (BatchNormalizatio… │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_add    │ (None, 32, 32,    │          0 │ conv2_block1_out… │
│ (Add)               │ 256)              │            │ conv2_block2_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block2_out    │ (None, 32, 32,    │          0 │ conv2_block2_add… │
│ (Activation)        │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_1_conv │ (None, 32, 32,    │     16,448 │ conv2_block2_out… │
│ (Conv2D)            │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_1_bn   │ (None, 32, 32,    │        256 │ conv2_block3_1_c… │
│ (BatchNormalizatio… │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_1_relu │ (None, 32, 32,    │          0 │ conv2_block3_1_b… │
│ (Activation)        │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_2_conv │ (None, 32, 32,    │     36,928 │ conv2_block3_1_r… │
│ (Conv2D)            │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_2_bn   │ (None, 32, 32,    │        256 │ conv2_block3_2_c… │
│ (BatchNormalizatio… │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_2_relu │ (None, 32, 32,    │          0 │ conv2_block3_2_b… │
│ (Activation)        │ 64)               │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_3_conv │ (None, 32, 32,    │     16,640 │ conv2_block3_2_r… │
│ (Conv2D)            │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_3_bn   │ (None, 32, 32,    │      1,024 │ conv2_block3_3_c… │
│ (BatchNormalizatio… │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_add    │ (None, 32, 32,    │          0 │ conv2_block2_out… │
│ (Add)               │ 256)              │            │ conv2_block3_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2_block3_out    │ (None, 32, 32,    │          0 │ conv2_block3_add… │
│ (Activation)        │ 256)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_1_conv │ (None, 16, 16,    │     32,896 │ conv2_block3_out… │
│ (Conv2D)            │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_1_bn   │ (None, 16, 16,    │        512 │ conv3_block1_1_c… │
│ (BatchNormalizatio… │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_1_relu │ (None, 16, 16,    │          0 │ conv3_block1_1_b… │
│ (Activation)        │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_2_conv │ (None, 16, 16,    │    147,584 │ conv3_block1_1_r… │
│ (Conv2D)            │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_2_bn   │ (None, 16, 16,    │        512 │ conv3_block1_2_c… │
│ (BatchNormalizatio… │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_2_relu │ (None, 16, 16,    │          0 │ conv3_block1_2_b… │
│ (Activation)        │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_0_conv │ (None, 16, 16,    │    131,584 │ conv2_block3_out… │
│ (Conv2D)            │ 512)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_3_conv │ (None, 16, 16,    │     66,048 │ conv3_block1_2_r… │
│ (Conv2D)            │ 512)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_0_bn   │ (None, 16, 16,    │      2,048 │ conv3_block1_0_c… │
│ (BatchNormalizatio… │ 512)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_3_bn   │ (None, 16, 16,    │      2,048 │ conv3_block1_3_c… │
│ (BatchNormalizatio… │ 512)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_add    │ (None, 16, 16,    │          0 │ conv3_block1_0_b… │
│ (Add)               │ 512)              │            │ conv3_block1_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block1_out    │ (None, 16, 16,    │          0 │ conv3_block1_add… │
│ (Activation)        │ 512)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_1_conv │ (None, 16, 16,    │     65,664 │ conv3_block1_out… │
│ (Conv2D)            │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_1_bn   │ (None, 16, 16,    │        512 │ conv3_block2_1_c… │
│ (BatchNormalizatio… │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_1_relu │ (None, 16, 16,    │          0 │ conv3_block2_1_b… │
│ (Activation)        │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_2_conv │ (None, 16, 16,    │    147,584 │ conv3_block2_1_r… │
│ (Conv2D)            │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_2_bn   │ (None, 16, 16,    │        512 │ conv3_block2_2_c… │
│ (BatchNormalizatio… │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_2_relu │ (None, 16, 16,    │          0 │ conv3_block2_2_b… │
│ (Activation)        │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_3_conv │ (None, 16, 16,    │     66,048 │ conv3_block2_2_r… │
│ (Conv2D)            │ 512)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_3_bn   │ (None, 16, 16,    │      2,048 │ conv3_block2_3_c… │
│ (BatchNormalizatio… │ 512)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_add    │ (None, 16, 16,    │          0 │ conv3_block1_out… │
│ (Add)               │ 512)              │            │ conv3_block2_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block2_out    │ (None, 16, 16,    │          0 │ conv3_block2_add… │
│ (Activation)        │ 512)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_1_conv │ (None, 16, 16,    │     65,664 │ conv3_block2_out… │
│ (Conv2D)            │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_1_bn   │ (None, 16, 16,    │        512 │ conv3_block3_1_c… │
│ (BatchNormalizatio… │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_1_relu │ (None, 16, 16,    │          0 │ conv3_block3_1_b… │
│ (Activation)        │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_2_conv │ (None, 16, 16,    │    147,584 │ conv3_block3_1_r… │
│ (Conv2D)            │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_2_bn   │ (None, 16, 16,    │        512 │ conv3_block3_2_c… │
│ (BatchNormalizatio… │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_2_relu │ (None, 16, 16,    │          0 │ conv3_block3_2_b… │
│ (Activation)        │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_3_conv │ (None, 16, 16,    │     66,048 │ conv3_block3_2_r… │
│ (Conv2D)            │ 512)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_3_bn   │ (None, 16, 16,    │      2,048 │ conv3_block3_3_c… │
│ (BatchNormalizatio… │ 512)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_add    │ (None, 16, 16,    │          0 │ conv3_block2_out… │
│ (Add)               │ 512)              │            │ conv3_block3_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block3_out    │ (None, 16, 16,    │          0 │ conv3_block3_add… │
│ (Activation)        │ 512)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_1_conv │ (None, 16, 16,    │     65,664 │ conv3_block3_out… │
│ (Conv2D)            │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_1_bn   │ (None, 16, 16,    │        512 │ conv3_block4_1_c… │
│ (BatchNormalizatio… │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_1_relu │ (None, 16, 16,    │          0 │ conv3_block4_1_b… │
│ (Activation)        │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_2_conv │ (None, 16, 16,    │    147,584 │ conv3_block4_1_r… │
│ (Conv2D)            │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_2_bn   │ (None, 16, 16,    │        512 │ conv3_block4_2_c… │
│ (BatchNormalizatio… │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_2_relu │ (None, 16, 16,    │          0 │ conv3_block4_2_b… │
│ (Activation)        │ 128)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_3_conv │ (None, 16, 16,    │     66,048 │ conv3_block4_2_r… │
│ (Conv2D)            │ 512)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_3_bn   │ (None, 16, 16,    │      2,048 │ conv3_block4_3_c… │
│ (BatchNormalizatio… │ 512)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_add    │ (None, 16, 16,    │          0 │ conv3_block3_out… │
│ (Add)               │ 512)              │            │ conv3_block4_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv3_block4_out    │ (None, 16, 16,    │          0 │ conv3_block4_add… │
│ (Activation)        │ 512)              │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_1_conv │ (None, 8, 8, 256) │    131,328 │ conv3_block4_out… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_1_bn   │ (None, 8, 8, 256) │      1,024 │ conv4_block1_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_1_relu │ (None, 8, 8, 256) │          0 │ conv4_block1_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_2_conv │ (None, 8, 8, 256) │    590,080 │ conv4_block1_1_r… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_2_bn   │ (None, 8, 8, 256) │      1,024 │ conv4_block1_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_2_relu │ (None, 8, 8, 256) │          0 │ conv4_block1_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_0_conv │ (None, 8, 8,      │    525,312 │ conv3_block4_out… │
│ (Conv2D)            │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_3_conv │ (None, 8, 8,      │    263,168 │ conv4_block1_2_r… │
│ (Conv2D)            │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_0_bn   │ (None, 8, 8,      │      4,096 │ conv4_block1_0_c… │
│ (BatchNormalizatio… │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_3_bn   │ (None, 8, 8,      │      4,096 │ conv4_block1_3_c… │
│ (BatchNormalizatio… │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_add    │ (None, 8, 8,      │          0 │ conv4_block1_0_b… │
│ (Add)               │ 1024)             │            │ conv4_block1_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block1_out    │ (None, 8, 8,      │          0 │ conv4_block1_add… │
│ (Activation)        │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_1_conv │ (None, 8, 8, 256) │    262,400 │ conv4_block1_out… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_1_bn   │ (None, 8, 8, 256) │      1,024 │ conv4_block2_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_1_relu │ (None, 8, 8, 256) │          0 │ conv4_block2_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_2_conv │ (None, 8, 8, 256) │    590,080 │ conv4_block2_1_r… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_2_bn   │ (None, 8, 8, 256) │      1,024 │ conv4_block2_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_2_relu │ (None, 8, 8, 256) │          0 │ conv4_block2_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_3_conv │ (None, 8, 8,      │    263,168 │ conv4_block2_2_r… │
│ (Conv2D)            │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_3_bn   │ (None, 8, 8,      │      4,096 │ conv4_block2_3_c… │
│ (BatchNormalizatio… │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_add    │ (None, 8, 8,      │          0 │ conv4_block1_out… │
│ (Add)               │ 1024)             │            │ conv4_block2_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block2_out    │ (None, 8, 8,      │          0 │ conv4_block2_add… │
│ (Activation)        │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_1_conv │ (None, 8, 8, 256) │    262,400 │ conv4_block2_out… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_1_bn   │ (None, 8, 8, 256) │      1,024 │ conv4_block3_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_1_relu │ (None, 8, 8, 256) │          0 │ conv4_block3_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_2_conv │ (None, 8, 8, 256) │    590,080 │ conv4_block3_1_r… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_2_bn   │ (None, 8, 8, 256) │      1,024 │ conv4_block3_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_2_relu │ (None, 8, 8, 256) │          0 │ conv4_block3_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_3_conv │ (None, 8, 8,      │    263,168 │ conv4_block3_2_r… │
│ (Conv2D)            │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_3_bn   │ (None, 8, 8,      │      4,096 │ conv4_block3_3_c… │
│ (BatchNormalizatio… │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_add    │ (None, 8, 8,      │          0 │ conv4_block2_out… │
│ (Add)               │ 1024)             │            │ conv4_block3_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block3_out    │ (None, 8, 8,      │          0 │ conv4_block3_add… │
│ (Activation)        │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_1_conv │ (None, 8, 8, 256) │    262,400 │ conv4_block3_out… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_1_bn   │ (None, 8, 8, 256) │      1,024 │ conv4_block4_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_1_relu │ (None, 8, 8, 256) │          0 │ conv4_block4_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_2_conv │ (None, 8, 8, 256) │    590,080 │ conv4_block4_1_r… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_2_bn   │ (None, 8, 8, 256) │      1,024 │ conv4_block4_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_2_relu │ (None, 8, 8, 256) │          0 │ conv4_block4_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_3_conv │ (None, 8, 8,      │    263,168 │ conv4_block4_2_r… │
│ (Conv2D)            │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_3_bn   │ (None, 8, 8,      │      4,096 │ conv4_block4_3_c… │
│ (BatchNormalizatio… │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_add    │ (None, 8, 8,      │          0 │ conv4_block3_out… │
│ (Add)               │ 1024)             │            │ conv4_block4_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block4_out    │ (None, 8, 8,      │          0 │ conv4_block4_add… │
│ (Activation)        │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_1_conv │ (None, 8, 8, 256) │    262,400 │ conv4_block4_out… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_1_bn   │ (None, 8, 8, 256) │      1,024 │ conv4_block5_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_1_relu │ (None, 8, 8, 256) │          0 │ conv4_block5_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_2_conv │ (None, 8, 8, 256) │    590,080 │ conv4_block5_1_r… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_2_bn   │ (None, 8, 8, 256) │      1,024 │ conv4_block5_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_2_relu │ (None, 8, 8, 256) │          0 │ conv4_block5_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_3_conv │ (None, 8, 8,      │    263,168 │ conv4_block5_2_r… │
│ (Conv2D)            │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_3_bn   │ (None, 8, 8,      │      4,096 │ conv4_block5_3_c… │
│ (BatchNormalizatio… │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_add    │ (None, 8, 8,      │          0 │ conv4_block4_out… │
│ (Add)               │ 1024)             │            │ conv4_block5_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block5_out    │ (None, 8, 8,      │          0 │ conv4_block5_add… │
│ (Activation)        │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_1_conv │ (None, 8, 8, 256) │    262,400 │ conv4_block5_out… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_1_bn   │ (None, 8, 8, 256) │      1,024 │ conv4_block6_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_1_relu │ (None, 8, 8, 256) │          0 │ conv4_block6_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_2_conv │ (None, 8, 8, 256) │    590,080 │ conv4_block6_1_r… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_2_bn   │ (None, 8, 8, 256) │      1,024 │ conv4_block6_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_2_relu │ (None, 8, 8, 256) │          0 │ conv4_block6_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_3_conv │ (None, 8, 8,      │    263,168 │ conv4_block6_2_r… │
│ (Conv2D)            │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_3_bn   │ (None, 8, 8,      │      4,096 │ conv4_block6_3_c… │
│ (BatchNormalizatio… │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_add    │ (None, 8, 8,      │          0 │ conv4_block5_out… │
│ (Add)               │ 1024)             │            │ conv4_block6_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv4_block6_out    │ (None, 8, 8,      │          0 │ conv4_block6_add… │
│ (Activation)        │ 1024)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_1_conv │ (None, 4, 4, 512) │    524,800 │ conv4_block6_out… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_1_bn   │ (None, 4, 4, 512) │      2,048 │ conv5_block1_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_1_relu │ (None, 4, 4, 512) │          0 │ conv5_block1_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_2_conv │ (None, 4, 4, 512) │  2,359,808 │ conv5_block1_1_r… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_2_bn   │ (None, 4, 4, 512) │      2,048 │ conv5_block1_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_2_relu │ (None, 4, 4, 512) │          0 │ conv5_block1_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_0_conv │ (None, 4, 4,      │  2,099,200 │ conv4_block6_out… │
│ (Conv2D)            │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_3_conv │ (None, 4, 4,      │  1,050,624 │ conv5_block1_2_r… │
│ (Conv2D)            │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_0_bn   │ (None, 4, 4,      │      8,192 │ conv5_block1_0_c… │
│ (BatchNormalizatio… │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_3_bn   │ (None, 4, 4,      │      8,192 │ conv5_block1_3_c… │
│ (BatchNormalizatio… │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_add    │ (None, 4, 4,      │          0 │ conv5_block1_0_b… │
│ (Add)               │ 2048)             │            │ conv5_block1_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block1_out    │ (None, 4, 4,      │          0 │ conv5_block1_add… │
│ (Activation)        │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_1_conv │ (None, 4, 4, 512) │  1,049,088 │ conv5_block1_out… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_1_bn   │ (None, 4, 4, 512) │      2,048 │ conv5_block2_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_1_relu │ (None, 4, 4, 512) │          0 │ conv5_block2_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_2_conv │ (None, 4, 4, 512) │  2,359,808 │ conv5_block2_1_r… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_2_bn   │ (None, 4, 4, 512) │      2,048 │ conv5_block2_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_2_relu │ (None, 4, 4, 512) │          0 │ conv5_block2_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_3_conv │ (None, 4, 4,      │  1,050,624 │ conv5_block2_2_r… │
│ (Conv2D)            │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_3_bn   │ (None, 4, 4,      │      8,192 │ conv5_block2_3_c… │
│ (BatchNormalizatio… │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_add    │ (None, 4, 4,      │          0 │ conv5_block1_out… │
│ (Add)               │ 2048)             │            │ conv5_block2_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block2_out    │ (None, 4, 4,      │          0 │ conv5_block2_add… │
│ (Activation)        │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_1_conv │ (None, 4, 4, 512) │  1,049,088 │ conv5_block2_out… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_1_bn   │ (None, 4, 4, 512) │      2,048 │ conv5_block3_1_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_1_relu │ (None, 4, 4, 512) │          0 │ conv5_block3_1_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_2_conv │ (None, 4, 4, 512) │  2,359,808 │ conv5_block3_1_r… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_2_bn   │ (None, 4, 4, 512) │      2,048 │ conv5_block3_2_c… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_2_relu │ (None, 4, 4, 512) │          0 │ conv5_block3_2_b… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_3_conv │ (None, 4, 4,      │  1,050,624 │ conv5_block3_2_r… │
│ (Conv2D)            │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_3_bn   │ (None, 4, 4,      │      8,192 │ conv5_block3_3_c… │
│ (BatchNormalizatio… │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ input_layer_68      │ (None, 35)        │          0 │ -                 │
│ (InputLayer)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_add    │ (None, 4, 4,      │          0 │ conv5_block2_out… │
│ (Add)               │ 2048)             │            │ conv5_block3_3_b… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ dense_220 (Dense)   │ (None, 512)       │     18,432 │ input_layer_68[0… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv5_block3_out    │ (None, 4, 4,      │          0 │ conv5_block3_add… │
│ (Activation)        │ 2048)             │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ dense_221 (Dense)   │ (None, 1024)      │    525,312 │ dense_220[0][0]   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ conv2d_49 (Conv2D)  │ (None, 1, 1, 128) │  4,194,432 │ conv5_block3_out… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ dense_222 (Dense)   │ (None, 2048)      │  2,099,200 │ dense_221[0][0]   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ flatten_33          │ (None, 128)       │          0 │ conv2d_49[0][0]   │
│ (Flatten)           │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ dense_223 (Dense)   │ (None, 512)       │  1,049,088 │ dense_222[0][0]   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ concatenate_33      │ (None, 640)       │          0 │ flatten_33[0][0], │
│ (Concatenate)       │                   │            │ dense_223[0][0]   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ dense_224 (Dense)   │ (None, 512)       │    328,192 │ concatenate_33[0… │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ dense_225 (Dense)   │ (None, 1024)      │    525,312 │ dense_224[0][0]   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ dense_226 (Dense)   │ (None, 2048)      │  2,099,200 │ dense_225[0][0]   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ dense_227 (Dense)   │ (None, 2048)      │  4,196,352 │ dense_226[0][0]   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ dense_228 (Dense)   │ (None, 512)       │  1,049,088 │ dense_227[0][0]   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ dense_229 (Dense)   │ (None, 1)         │        513 │ dense_228[0][0]   │
└─────────────────────┴───────────────────┴────────────┴───────────────────┘

 Total params: 71,843,077 (274.06 MB)

 Trainable params: 16,085,121 (61.36 MB)

 Non-trainable params: 23,587,712 (89.98 MB)

 Optimizer params: 32,170,244 (122.72 MB)

None
Out[85]:
No description has been provided for this image

Transfer Learning : ResNet #3¶

In [100]:
adam_optimizer = Adam(learning_rate=0.0075, beta_1=0.9, beta_2=0.999, epsilon=1e-07)
sgd_optimizer = SGD(learning_rate=1e-4 , momentum=0.9, nesterov=True)

filepath = './dnn/best_weights-resNet3.keras'
checkpointer = ModelCheckpoint(filepath=filepath, verbose=0, save_best_only=True)
monitor = EarlyStopping(monitor='val_loss', min_delta=1e-3, patience=25, verbose=1, mode='auto')
    
resNet = ResNet50(weights='imagenet', include_top=False, input_shape=(128, 128, 3)) 

for layer in resNet.layers:
    layer.trainable = True

conv_layer = Conv2D(128, kernel_size=4, activation='relu')(resNet.output)  
flatten_res = Flatten()(conv_layer)  


visible1 = resNet.input

# Tabular data input model
visible2 = Input(shape=(x.shape[1],))
dense1 = Dense(512, activation='relu')(visible2)
dense2 = Dense(1024, activation='relu')(dense1)
dense3 = Dense(2048, activation='relu')(dense2)
# dense3 = Dropout(0.1)(dense3)
dense4 = Dense(512, activation='relu')(dense3)

merge = concatenate([flatten_res, dense4])

hidden1 = Dense(512, activation='relu')(merge)
hidden2 = Dense(1024, activation='relu')(hidden1)
hidden3 = Dense(2048, activation='relu')(hidden2)
# hidden3 = Dropout(0.2)(hidden3)
hidden4 = Dense(2048, activation='relu')(hidden3)
hidden5 = Dense(512, activation='relu')(hidden4)
output = Dense(1)(hidden5)

model = Model(inputs=[visible1, visible2], outputs=output)

# print(model.summary())
model.compile(loss='mean_squared_error', optimizer=adam_optimizer)
model.fit([x_train_img, x_train], y_train,
            validation_data=([x_test_img, x_test], y_test),
            callbacks=[monitor, checkpointer],verbose=2,epochs=150)

print('Training finished...Loading the best model')
print()
model.load_weights(filepath) # load weights from best model

pred_resNet3 = model.predict([x_test_img, x_test])
score_resNet3 = np.sqrt(metrics.mean_squared_error(pred_resNet3,y_test))
print("Score (RMSE): {}".format(score_resNet3))

Epoch 1/150
8/8 - 81s - 10s/step - loss: 21457895686144.0000 - val_loss: inf
Epoch 2/150
8/8 - 1s - 129ms/step - loss: 334170357760.0000 - val_loss: inf
Epoch 3/150
8/8 - 1s - 116ms/step - loss: 207081947136.0000 - val_loss: inf
Epoch 4/150
8/8 - 3s - 409ms/step - loss: 168297611264.0000 - val_loss: 9808948503951855215782212646469632.0000
Epoch 5/150
8/8 - 5s - 590ms/step - loss: 110809612288.0000 - val_loss: 11863660548948005796769840496640.0000
Epoch 6/150
8/8 - 4s - 542ms/step - loss: 91185160192.0000 - val_loss: 52509409632972446024649932800.0000
Epoch 7/150
8/8 - 3s - 427ms/step - loss: 51178123264.0000 - val_loss: 510849004047171880621178880.0000
Epoch 8/150
8/8 - 3s - 426ms/step - loss: 27910414336.0000 - val_loss: 7019695166771817600778240.0000
Epoch 9/150
8/8 - 4s - 501ms/step - loss: 21446799360.0000 - val_loss: 192468013417525164376064.0000
Epoch 10/150
8/8 - 3s - 419ms/step - loss: 13124233216.0000 - val_loss: 4390464982606298480640.0000
Epoch 11/150
8/8 - 4s - 467ms/step - loss: 9879970816.0000 - val_loss: 52687576856235343872.0000
Epoch 12/150
8/8 - 4s - 548ms/step - loss: 9667404800.0000 - val_loss: 1676710702137475072.0000
Epoch 13/150
8/8 - 3s - 412ms/step - loss: 9983817728.0000 - val_loss: 36128839692189696.0000
Epoch 14/150
8/8 - 4s - 457ms/step - loss: 5374980096.0000 - val_loss: 2002834652397568.0000
Epoch 15/150
8/8 - 4s - 486ms/step - loss: 4259698176.0000 - val_loss: 25525296824320.0000
Epoch 16/150
8/8 - 3s - 414ms/step - loss: 3630961408.0000 - val_loss: 31788214272.0000
Epoch 17/150
8/8 - 1s - 127ms/step - loss: 3633146368.0000 - val_loss: 56455737344.0000
Epoch 18/150
8/8 - 1s - 118ms/step - loss: 5066958336.0000 - val_loss: 83259056128.0000
Epoch 19/150
8/8 - 1s - 118ms/step - loss: 3359559424.0000 - val_loss: 82531827712.0000
Epoch 20/150
8/8 - 1s - 118ms/step - loss: 3628035328.0000 - val_loss: 67713249280.0000
Epoch 21/150
8/8 - 1s - 119ms/step - loss: 2291453440.0000 - val_loss: 78411513856.0000
Epoch 22/150
8/8 - 1s - 118ms/step - loss: 1577149568.0000 - val_loss: 70351028224.0000
Epoch 23/150
8/8 - 1s - 118ms/step - loss: 2162359552.0000 - val_loss: 64758988800.0000
Epoch 24/150
8/8 - 1s - 118ms/step - loss: 2188599552.0000 - val_loss: 67543506944.0000
Epoch 25/150
8/8 - 1s - 118ms/step - loss: 1397987200.0000 - val_loss: 72141430784.0000
Epoch 26/150
8/8 - 1s - 118ms/step - loss: 1897369856.0000 - val_loss: 57386754048.0000
Epoch 27/150
8/8 - 1s - 119ms/step - loss: 1355569024.0000 - val_loss: 43981090816.0000
Epoch 28/150
8/8 - 1s - 119ms/step - loss: 3084919296.0000 - val_loss: 62863466496.0000
Epoch 29/150
8/8 - 1s - 119ms/step - loss: 2202707200.0000 - val_loss: 65443311616.0000
Epoch 30/150
8/8 - 1s - 118ms/step - loss: 1487263616.0000 - val_loss: 52634767360.0000
Epoch 31/150
8/8 - 1s - 118ms/step - loss: 1208833408.0000 - val_loss: 54866755584.0000
Epoch 32/150
8/8 - 1s - 119ms/step - loss: 1100148736.0000 - val_loss: 51140866048.0000
Epoch 33/150
8/8 - 1s - 120ms/step - loss: 1479528192.0000 - val_loss: 57578721280.0000
Epoch 34/150
8/8 - 1s - 119ms/step - loss: 1136723456.0000 - val_loss: 52306075648.0000
Epoch 35/150
8/8 - 1s - 119ms/step - loss: 1494099584.0000 - val_loss: 54254084096.0000
Epoch 36/150
8/8 - 1s - 119ms/step - loss: 738546368.0000 - val_loss: 52105428992.0000
Epoch 37/150
8/8 - 1s - 120ms/step - loss: 1321481728.0000 - val_loss: 54827958272.0000
Epoch 38/150
8/8 - 1s - 119ms/step - loss: 1011561344.0000 - val_loss: 46396653568.0000
Epoch 39/150
8/8 - 1s - 119ms/step - loss: 832796672.0000 - val_loss: 52856401920.0000
Epoch 40/150
8/8 - 1s - 119ms/step - loss: 1141399168.0000 - val_loss: 49522290688.0000
Epoch 41/150
8/8 - 1s - 121ms/step - loss: 1342953472.0000 - val_loss: 60634574848.0000
Epoch 41: early stopping
Training finished...Loading the best model

2/2 ━━━━━━━━━━━━━━━━━━━━ 7s 3s/step
Score (RMSE): 178292.515625